# Clunat - Complete Research, Architecture, and Governance Knowledge Base > Clunat conducts fundamental research into autonomous intelligence, sovereign AI infrastructure, and foundational intelligence systems. Canonical Domain: https://clunat.com Standard: LLMs Full Context Documentation (llms-full.txt) -------------------------------------------------------------------------------- ## Index Slug: /research-index Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Index Slug: /research-overview Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Index Slug: /research-reasoning Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Index Slug: /research-architectures Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Index Slug: /research-ai-safety Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## About Clunat Slug: /company-about Summary: Intelligence for Human Flourishing Last Updated: 2026 ### We believe artificial general intelligence should help humanity learn, create, innovate, and grow safely. Clunat is an independent artificial intelligence and fundamental research laboratory dedicated to advancing cognitive systems, foundational reasoning architectures, and beneficial machine intelligence. We explore the foundations of increasingly capable AI systems, from how machines represent knowledge and reason over complex problems to how intelligence can become more efficient, general, reliable, and controllable. Our work sits at the intersection of artificial intelligence, mathematics, computer science, scientific computing, cognitive architectures, and AI safety. We pursue fundamental research not simply to build larger systems, but to understand the principles that make intelligence possible and to translate those principles into systems that can create meaningful value for humanity. "The goal of intelligence is not intelligence for its own sake. The goal is to expand what humanity can understand, create, discover, and achieve." ### Our Mission Our mission is to develop increasingly capable intelligence while advancing the scientific foundations required to make that intelligence useful, reliable, efficient, and safe. We are interested in the difficult questions underlying advanced AI: - How can a machine reason rather than merely predict? - How can an AI system construct, revise, and verify its own solutions? - How can intelligence become more computationally efficient without sacrificing capability? - How can models acquire broader and more transferable forms of reasoning? - How can an AI system understand uncertainty and recognize the limits of its own knowledge? - How can advanced intelligence remain interpretable, controllable, and aligned with human objectives as its capabilities increase? ### The Central Question "How can we build intelligence that expands the frontier of human possibility without compromising human agency?" ### The Four Pillars of Clunat 1. Learn We want AI to become a powerful instrument for human understanding. Clunat researches systems capable of synthesizing knowledge across disciplines, reasoning through complex problems, identifying relationships between ideas, and assisting humans in navigating increasingly large bodies of information. A truly useful intelligence should not merely tell a person what is already known. It should help them understand why something is true, explore competing explanations, identify missing information, connect concepts across domains, and construct new hypotheses. Our objective is not to replace human curiosity: it is to multiply it. 2. Create Intelligence should expand the creative potential of people. Clunat develops generative and reasoning technologies intended to augment researchers, developers, engineers, designers, writers, scientists, and creators. We believe the most powerful creative systems will not simply generate content on command. They will participate in an iterative process of thinking, exploring, critiquing, refining, and building alongside humans as cognitive collaborators. 3. Innovate The most important advances in artificial intelligence may emerge from solving problems that are not visible from the surface of today's systems. Clunat invests in fundamental research into the underlying mechanisms of intelligence: reasoning architectures, efficient attention, neural compression, memory, adaptive computation, and machine-assisted scientific discovery. Building increasingly capable systems cannot depend indefinitely on simply increasing computational scale. Intelligence must also become fundamentally more efficient in how it represents information, allocates computation, retrieves knowledge, and learns from experience. 4. Grow Safely Capability without safety is not progress. The development of increasingly powerful AI systems must be accompanied by an equally serious effort to understand, evaluate, and control those systems. Safety is not a layer added after research is complete: it is part of the research itself. Our approach combines alignment research, mechanistic investigation, adversarial evaluation, capability characterization, security engineering, and rigorous institutional governance. ### Our Research Philosophy At Clunat, we approach advanced AI as both a scientific problem and a societal responsibility. Ambitious research and responsible development are compatible: - Scientific exploration should remain bold. - Engineering should remain rigorous. - Evaluation should remain skeptical. - Governance should remain accountable. - Safety should remain inseparable from capability. ### Fundamental Research First Our foundation is research. We investigate underlying mechanisms that contribute to more capable forms of machine intelligence: - Representation: How intelligent systems encode information, concepts, and abstractions. - Reasoning: How systems perform structured inference over interconnected steps. - Memory: How intelligence retains, retrieves, and updates information efficiently. - Computation: Dynamically allocating compute to steps requiring deeper reasoning. - Efficiency: Operating with significantly lower computational and energy requirements. - Autonomy: Executing complex goals while bounded by explicit human permissions. - Safety: Understanding and constraining advanced systems before capabilities exceed evaluation. ### Building for a Global Scientific Community Clunat is proudly based in South Asia, India, while its research ambition is global. Scientific progress does not belong to a single country, institution, or company. We want to contribute to a broader ecosystem of researchers, engineers, students, developers, universities, laboratories, and independent thinkers working toward a deeper understanding of intelligence. Geography should not determine who gets to participate in the future of AI. ### The Clunat Standard "Do not build merely because something can be built. Build because it expands meaningful human capability, and build it with enough scientific and engineering discipline to understand its consequences." Progress occurs when capability is accompanied by understanding. Progress occurs when systems become more useful without becoming unnecessarily dangerous, when scientific knowledge becomes accessible, and when humanity gains new tools to solve problems previously beyond reach. Building intelligence for discovery, creation, and human flourishing. -------------------------------------------------------------------------------- ## The Four Pillars of Clunat Slug: /company-pillars Summary: Learn, Create, Innovate, and Grow Safely. Last Updated: 2026 ### 1. Learn We want AI to become a powerful instrument for human understanding. Clunat researches systems capable of synthesizing knowledge across disciplines, reasoning through complex problems, identifying relationships between ideas, and assisting humans in navigating increasingly large bodies of information. A truly useful intelligence should not merely tell a person what is already known. It should help them understand why something is true, explore competing explanations, identify missing information, connect concepts across domains, and construct new hypotheses. Our objective is not to replace human curiosity: it is to multiply it. ### 2. Create Intelligence should expand the creative potential of people. Clunat develops generative and reasoning technologies intended to augment researchers, developers, engineers, designers, writers, scientists, and creators. We believe the most powerful creative systems will not simply generate content on command. They will participate in an iterative process of thinking, exploring, critiquing, refining, and building alongside humans as cognitive collaborators. ### 3. Innovate The most important advances in artificial intelligence may emerge from solving problems that are not visible from the surface of today's systems. Clunat invests in fundamental research into the underlying mechanisms of intelligence: reasoning architectures, efficient attention, neural compression, memory, adaptive computation, and machine-assisted scientific discovery. Building increasingly capable systems cannot depend indefinitely on simply increasing computational scale. Intelligence must also become fundamentally more efficient in how it represents information, allocates computation, retrieves knowledge, and learns from experience. ### 4. Grow Safely Capability without safety is not progress. The development of increasingly powerful AI systems must be accompanied by an equally serious effort to understand, evaluate, and control those systems. Safety is not a layer added after research is complete: it is part of the research itself. Our approach combines alignment research, mechanistic investigation, adversarial evaluation, capability characterization, security engineering, and rigorous institutional governance. -------------------------------------------------------------------------------- ## Our Research Philosophy Slug: /company-philosophy Summary: Approaching advanced AI as both a scientific problem and a societal responsibility. Last Updated: 2026 ### Research Principles At Clunat, we approach advanced AI as both a scientific problem and a societal responsibility. Ambitious research and responsible development are compatible: - Scientific exploration should remain bold. - Engineering should remain rigorous. - Evaluation should remain skeptical. - Governance should remain accountable. - Safety should remain inseparable from capability. ### Fundamental Research First Our foundation is research. We investigate underlying mechanisms that contribute to more capable forms of machine intelligence: - Representation: How intelligent systems encode information, concepts, and abstractions. - Reasoning: How systems perform structured inference over interconnected steps. - Memory: How intelligence retains, retrieves, and updates information efficiently. - Computation: Dynamically allocating compute to steps requiring deeper reasoning. - Efficiency: Operating with significantly lower computational and energy requirements. - Autonomy: Executing complex goals while bounded by explicit human permissions. - Safety: Understanding and constraining advanced systems before capabilities exceed evaluation. ### The Clunat Standard "Do not build merely because something can be built. Build because it expands meaningful human capability, and build it with enough scientific and engineering discipline to understand its consequences." Progress occurs when capability is accompanied by understanding. Progress occurs when systems become more useful without becoming unnecessarily dangerous, when scientific knowledge becomes accessible, and when humanity gains new tools to solve problems previously beyond reach. Building intelligence for discovery, creation, and human flourishing. -------------------------------------------------------------------------------- ## Partner Network Services Slug: /partner-network Summary: A global alliance of certified system integrators, technical consultancies, and independent software vendors deploying frontier Clunat cognitive architectures at enterprise scale. Last Updated: 2026 ### Accelerating Safe Enterprise AI Adoption *Bridging frontier foundation models with mission-critical corporate operations* The Clunat Partner Network (CPN) connects leading global system integrators, independent software vendors (ISVs), and domain-specialized consultancies with the advanced cognitive infrastructure of Clunat. As organizations transition from experimental pilots to production-grade agentic systems, our partner network provides the strategic advisory, custom model adaptation, sovereign data isolation, and aligned AI safety engineering necessary for dependable real-world execution. - Direct technical co-engineering and architectural reviews with Clunat foundation researchers - Priority access to unreleased model weights, high-concurrency inference endpoints, and specialized fine-tuning toolchains - Formal Aligned AI audit, verification, and red-teaming certification programs - Co-selling enablement, shared enterprise pipelines, and dedicated executive partner sponsorship ### Partnership Tracks *Tailored programs engineered for distinct technical and commercial collaboration models* We offer three specialized partnership tracks to support diverse organizational capabilities and client engagements: #### Global System Integrators & Consultancies [Advisory & Systems Integration] For enterprise advisory firms and digital transformation leaders delivering multi-agent workflows, legacy ERP integrations, and sovereign on-premises deployments. #### Independent Software Vendors (ISVs) [Product Integration] For SaaS providers and product teams embedding Clunat foundation reasoning, automated code generation, and multi-modal intelligence into commercial platforms. #### Applied Engineering & Research Boutiques [R&D Collaboration] For specialized AI engineering labs and academic centers co-developing domain-specific cognitive benchmarks, automated safety monitors, and specialized tool-use pipelines. ### Partner Enablement & Infrastructure *Tools, sandboxes, and engineering resources provided to verified partners* Verified partners operate within an accelerated technical environment designed to eliminate deployment friction and ensure predictable performance under stringent enterprise conditions. - Dedicated Developer Sandboxes: High-throughput, isolated environments for latency profiling, load testing, and edge deployment validation. - Direct Engineering Channels: Direct communication channels with Clunat core infrastructure leads and bi-weekly technical syncs. - Guaranteed Enterprise SLAs: Priority model routing, reserved GPU inference clusters, and guaranteed uptime commitments for mission-critical workloads. - Joint Go-To-Market Programs: Co-branded whitepapers, customer case studies, and presentation opportunities at global industry conferences. ### Quality, Safety & Certification Standards *Upholding aligned AI safety and ethical governance across all client deliverables* Every member of the Clunat Partner Network commits to our core Safety Policy and Responsible Scaling Framework. System architectures must satisfy rigorous audit procedures before receiving certification. - Zero-Retention Guarantee: Customer data submitted through partner implementations is never logged or utilized for base foundation model training. - Aligned AI Red-Teaming: Standardized automated safety harnesses to audit prompt boundaries, bias, and output veracity prior to production sign-off. - Annual Technical Recertification: Continuous credentialing to ensure partner engineering teams remain proficient in the latest Clunat model releases. ### Join the Partner Network *Submit your organization credentials to begin the accreditation review* Whether your organization is seeking to become a certified implementation partner or looking to engage an accredited Clunat integrator for your enterprise deployment, our partner alliances team is here to assist. Partner Alliances: partners@clunat.com | Developer Relations: dev@clunat.com | Official Portal: platform.clunat.com -------------------------------------------------------------------------------- ## Contact Slug: /company-contact Summary: Reach out to our team for enterprise partnerships, research collaborations, or press inquiries. Last Updated: 2026 ### Send Us a Message *We respond to inquiries within 24 business hours* Whether you are looking to integrate Clunat into enterprise infrastructure, build on our developer platform, or discuss research collaborations, our team is here to assist. Official Website: clunat.com | Contact: contect@clunat.com | Developers: dev@clunat.com | Research: research@clunat.com -------------------------------------------------------------------------------- ## Governance Slug: /governance Summary: Status: Internal governance baseline for Clunat AI, AGI-oriented research, and products. Last Updated: 2026 ### 1. Purpose and Scope Clunat develops artificial intelligence, advanced AI systems, AGI-oriented research, and scientific technologies. This Governance Framework establishes the internal rules by which Clunat makes consequential decisions about research, models, products, data, security, deployment, partnerships, and organizational risk. It is intended to function as an operating system for responsible decision-making, not as a statement of aspirations alone. The framework applies across the company, including research, engineering, product, security, data, infrastructure, operations, commercial activity, and leadership. It covers the full AI lifecycle: problem selection, research, data acquisition, training, evaluation, red-teaming, deployment, monitoring, incident response, modification, retirement, and post-deployment learning. Clunat should treat governance as a technical and organizational capability. NIST's AI Risk Management Framework describes governance as a cross-cutting function that should be integrated throughout the AI lifecycle, while the OECD AI Principles emphasize accountability, transparency, robustness, security, safety, and systematic risk management. This framework adapts those ideas to an AI/AGI research company and adds stronger internal controls for advanced-model development. ### 2. Governance Principles Clunat adopts eight governing principles: - First, mission before momentum: Commercial pressure, benchmark competition, fundraising objectives, or publication incentives must not override a material safety, security, legal, or integrity concern. - Second, accountable ownership: Every consequential system, experiment, dataset, model release, and external commitment must have an identifiable owner with authority to act and a documented decision trail. - Third, independent challenge: Teams responsible for building a system should not be the only people deciding whether that system is safe or ready for release. Where risk is material, an independent evaluation function must have authority to block or delay deployment. - Fourth, proportionality: Governance should become stricter as capability, autonomy, scale, access, and potential harm increase. Low-risk experimentation should not carry frontier-model bureaucracy, but frontier systems should never be governed like ordinary software. - Fifth, evidence over intuition: Release and risk decisions should rely on evaluations, reproducible measurements, incident evidence, threat models, and documented uncertainty. - Sixth, human oversight and reversibility: Systems should have mechanisms for intervention, access restriction, rollback, shutdown, and safe decommissioning appropriate to their risk. - Seventh, scientific integrity: Clunat should preserve accurate records, disclose meaningful limitations, distinguish established findings from hypotheses, and avoid manipulating evaluations or selectively reporting results. - Eighth, continuous improvement: Governance is not a one-time approval. Policies, thresholds, evaluations, and organizational controls must evolve with evidence, capability changes, and external developments. ### 3. Governance Architecture The governance architecture consists of five layers: - The Board or Equivalent Governing Body: Provides fiduciary oversight, approves the company's risk appetite, oversees the most consequential AI risks, and receives periodic reporting on safety, security, legal exposure, major incidents, and strategic research risks. - The Executive Leadership Team: Converts board-level policy into operating priorities and is accountable for implementation. The Chief Executive or equivalent executive retains overall responsibility for organizational execution, but may not unilaterally override a formally established safety stop without documenting the rationale and obtaining the required governance review. - The AI Governance and Risk Committee: The central cross-functional decision body for high-impact AI matters. It includes leadership from research, engineering, security, legal/compliance, product, and safety/evaluation. Its responsibilities include model release gates, risk acceptance, high-severity incidents, exceptional research approvals, and material changes to safety controls. - The Independent Evaluation and Safety Function: Performs adversarial testing, capability evaluations, misuse testing, robustness analysis, and release-readiness reviews. Its independence is protected from product launch incentives. It maintains a direct escalation path to the AI Governance and Risk Committee and, for critical issues, to the board. - Research and Product Teams: Remain responsible for implementation and first-line risk management. Governance is therefore distributed: builders own day-to-day controls, independent evaluators challenge assumptions, and senior governance bodies make decisions on material residual risk. ### 4. Decision Rights and the Three-Lines Model Clunat operates a three-lines model: Line 1: Build and Operate: Research, engineering, product, infrastructure, and operations identify and manage risks in normal work. Each project maintains an owner, system description, risk register, evaluation plan, and change history. Line 2: Risk, Safety, Security, and Compliance: Independent functions define standards, review risk assessments, conduct challenge, monitor compliance, and escalate unresolved concerns. They have authority to require additional testing or controls. Line 3: Assurance: Periodic internal review or independent external assurance examines whether governance processes actually work. Assurance tests records, approvals, evaluation integrity, incident handling, access controls, and exceptions rather than simply checking whether policies exist. Decision rights must be explicit. A research lead may approve routine experiments within predefined boundaries. A release owner may approve low-risk releases that satisfy standard gates. High-impact or frontier-capability releases require committee approval. A critical safety or security issue can trigger an immediate deployment freeze without waiting for commercial approval. ### 5. AI Research Governance Research freedom is essential to Clunat, but research involving advanced capabilities requires structured controls. Every major research program should have a research charter specifying its objective, expected capabilities, known risks, datasets or tools involved, compute requirements, responsible owner, and intended publication or deployment path. Research should be classified by capability and risk. Categories include routine research, elevated-capability research, advanced capability research, and frontier/AGI-relevant research. The higher categories trigger stronger access controls, independent review, restricted environments, additional logging, and pre-defined stop conditions. Particular attention is given to research that could materially increase autonomous operation, cyber capability, biological or chemical assistance, large-scale persuasion, self-replication, automated exploitation, strategic deception, model autonomy, or the ability of systems to circumvent human oversight. The purpose is not to prohibit difficult science; it is to ensure that capability development and risk understanding progress together. Research records preserve hypotheses, experiment configurations, model versions, evaluation scripts, relevant failures, and material negative results. Reproducibility and traceability are governance assets because they allow the company to determine what changed and why a system behaves differently. ### 6. Model Development and Release Gates Every significant model passes a staged release process: Gate 0: Project Authorization: Define scope, owner, threat model, data provenance expectations, and evaluation plan. Gate 1: Development Readiness: Establish secure infrastructure, logging, access controls, and baseline evaluations. Gate 2: Capability Characterization: Measure intended and unintended capabilities, including meaningful uncertainty. Gate 3: Adversarial Evaluation: Conduct red-teaming, misuse testing, robustness testing, security testing, and targeted evaluations for the model's risk profile. Gate 4: Deployment Approval: Document residual risk, mitigations, access model, monitoring, rollback plan, and accountable approvers. Gate 5: Post-Deployment Review: Monitor incidents, drift, misuse, and unexpected capabilities to determine whether controls remain adequate. No model should be released merely because it performs well on benchmarks. Benchmark performance is evidence about capability, not proof of safety. Evaluation includes failure modes, distribution shift, adversarial prompting, tool use, autonomy, and system-level behavior where applicable. For advanced models, Clunat maintains explicit release thresholds. A threshold can be capability-based, risk-based, or both. If a model crosses a predefined capability boundary, the release process automatically escalates to enhanced governance even if the model is commercially attractive. ### 7. AGI and Frontier-System Governance AGI-oriented research requires a higher governance tier because the uncertainty around generality, autonomy, emergent capabilities, and downstream effects can be greater than for conventional AI products. Clunat establishes a Frontier Systems Protocol. Under this protocol, a system may be designated frontier when its capabilities, autonomy, scale, tool access, or potential impact exceed internally defined thresholds. Frontier designation triggers enhanced compute and infrastructure security, restricted access, independent evaluations, senior governance review, documented deployment assumptions, and continuous monitoring. Clunat avoids a single-number definition of AGI. Instead, governance evaluates a capability vector: breadth of tasks, reasoning performance, autonomy, persistence, tool use, ability to plan over long horizons, ability to modify or generate software, interaction with external systems, and evidence of novel or unexpected capabilities. For frontier systems, the company maintains a capability and risk dossier. It includes evaluation methodology, limitations, known failure modes, dangerous capability testing where appropriate, mitigations, residual risks, access restrictions, incident procedures, and release rationale. Where evaluations are incomplete, the uncertainty itself is recorded as a risk factor. ### 8. Safety, Security, and Misuse Prevention Safety and security are separate but interacting disciplines. AI safety addresses harmful behavior and system-level risk; security protects the model, infrastructure, weights, credentials, datasets, evaluation systems, and deployment environment from unauthorized access or manipulation. Clunat maintains threat models for major systems and updates them when capabilities or deployment conditions change. Critical assets use least-privilege access, strong authentication, secrets management, network segmentation, audit logging, secure software supply chains, and controlled model-weight access. Misuse controls are proportionate to the system. Depending on risk, controls include rate limits, user verification, content and tool-use policies, sandboxing, permission boundaries, human review, abuse monitoring, and restricted APIs. These controls are tested rather than assumed to work. Safety evaluation includes both ordinary and adversarial conditions. Red-team findings are tracked to closure, risk acceptance, or a documented decision not to remediate. Severe findings automatically trigger escalation and, where appropriate, suspension of deployment. ### 9. Data, Privacy, and Intellectual Property Governance Clunat maintains a documented data governance lifecycle covering acquisition, licensing, provenance, classification, storage, transformation, use, retention, and deletion. Training and evaluation datasets are traceable to their source and permitted use. Data that is not appropriate for a stated purpose does not enter the training pipeline merely because it is technically accessible. Metadata is separated from training content where practical, and datasets have machine-readable manifests describing origin, license or usage basis, preprocessing, exclusions, known limitations, and quality statistics. Data quality controls detect duplication, contamination, malicious content, personally identifiable information where relevant, and other unsuitable material. Privacy governance reflects applicable law and contractual commitments. Sensitive data receives stronger access controls and retention limits. Where personal data is processed, Clunat documents the purpose, lawful basis where applicable, safeguards, and deletion or correction mechanisms required by the applicable regime. Intellectual property is governed through documented ownership, licensing, contribution records, and review of external materials. Researchers must not knowingly misrepresent third-party material as original Clunat work. Open-source and research publication practices remain compatible with the security and licensing posture of the system. ### 10. Incident Response and Escalation Clunat maintains a unified AI incident process covering safety incidents, security breaches, privacy events, harmful misuse, evaluation failures, material model regressions, and governance violations. Incidents are classified by severity. Critical incidents include credible risks of severe harm, major unauthorized access to sensitive model assets, dangerous capability exposure, significant loss of control, or failures that materially undermine a high-impact deployment. Critical incidents require immediate containment, executive notification, preservation of evidence, and governance review. The incident process follows a defined sequence: detect, contain, assess, escalate, remediate, communicate, and learn. Containment can include disabling an endpoint, revoking credentials, restricting a model, freezing a release, or isolating infrastructure. Post-incident reviews identify root causes rather than assigning blame alone. Corrective actions have assigned owners and deadlines. Material incidents feed back into evaluation suites, threat models, engineering controls, and governance thresholds. ### 11. Ethics, Conflicts, and Whistleblowing Clunat maintains a professional ethics standard covering honesty in research, conflicts of interest, misuse of company resources, confidentiality, harassment, retaliation, manipulation of evaluations, and improper influence over governance decisions. Conflicts of interest must be disclosed when personal, financial, academic, or external relationships could reasonably affect a decision. The appropriate response may include recusal, independent review, or additional approval. Employees and collaborators have a secure channel to report concerns. Whistleblowers are protected from retaliation when raising concerns in good faith. A governance system that technically contains escalation channels but discourages their use is not functional. Evaluation independence is strictly enforced. Researchers must not be pressured to change test definitions, remove inconvenient failures, or delay reporting solely to make a model appear safer or more capable than the evidence supports. ### 12. Transparency, Documentation, and Auditability Governance decisions must be reconstructable after the fact. For material decisions, Clunat retains the decision, date, responsible individuals, evidence considered, alternatives, residual risks, conditions imposed, and follow-up requirements. Model cards or equivalent system documentation communicate intended use, limitations, evaluation results, major risks, access restrictions, and known failure modes. Internal dossiers contain more granular information than public documentation. Transparency does not require publishing dangerous operational details. Clunat distinguishes between public transparency, stakeholder transparency, and restricted internal records. Sensitive security information may require controlled disclosure, but the existence of a decision and its accountability remain recorded. Periodic governance audits examine whether release gates are being followed, whether exceptions are accumulating, whether incidents are closed effectively, and whether risk thresholds remain aligned with actual capabilities. ### 13. Risk Appetite and Exception Management Clunat maintains a written risk appetite approved by senior governance. The company accepts ordinary engineering uncertainty, research uncertainty, and bounded commercial risk; it does not casually accept unknown or unbounded risk involving severe potential harm, loss of control, major security compromise, or deliberate falsification of evidence. Exceptions must be explicit, time-bounded, owned, and approved at the appropriate governance level. An exception is not a way to bypass governance permanently. It is a controlled departure from a requirement with a documented rationale and compensating controls. Repeated exceptions trigger policy review. If a team continuously requires an exception, the underlying process may be unrealistic, or the organization may be systematically underestimating risk. ### 14. Scaling Governance with the Company Governance scales in maturity as Clunat grows. At the earliest stage, the founder and leadership group combine roles, but core controls: ownership, model inventory, risk classification, evaluation records, incident response, and release gates, exist from the beginning. As the company grows, Clunat separates research leadership from independent safety/evaluation leadership, formalizes board-level risk oversight, introduces internal audit or external assurance, and establishes specialized committees for security, privacy, research integrity, and frontier AI when justified. Governance is reviewed at least annually and after major capability changes, significant incidents, new regulatory obligations, material acquisitions, or major shifts in operational scope. The framework operates as a living control system rather than a static document. ### 15. Minimum Governance Controls Clunat maintains, at minimum: a current AI system inventory; named owners for significant systems; risk classifications; model and dataset provenance records; documented evaluation plans; release approvals; security access controls; incident response procedures; a whistleblowing channel; conflict-of-interest disclosures; exception registers; material decision logs; and periodic governance reviews. For advanced or frontier systems, minimum controls additionally include independent evaluation, explicit capability thresholds, enhanced infrastructure security, restricted access, red-team testing, documented residual risk, rollback or shutdown mechanisms, and senior-level release authorization. These controls are operational. The objective is to make the safe path the normal path: researchers know what is required, reviewers know what evidence to expect, and leadership knows when a decision must be escalated. ### 16. Governance Charter and Approval This Framework is adopted as the baseline governance standard for Clunat AI, AGI-oriented research, and related products. It is accompanied by detailed procedures, technical standards, model release checklists, incident playbooks, data governance standards, security policies, and research integrity procedures. The framework operates as an internal governance standard. Applicable law, regulation, contractual obligations, and corporate governance documents take precedence where they impose stricter requirements. The central operating principle is direct: Clunat builds intelligence with the same discipline with which it builds infrastructure. Capability creates opportunity, but governance determines whether that capability can be developed and deployed responsibly. For an AI/AGI research company, governance is part of the research and engineering system itself. -------------------------------------------------------------------------------- ## Governance Slug: /governance-framework Summary: Status: Internal governance baseline for Clunat AI, AGI-oriented research, and products. Last Updated: 2026 ### 1. Purpose and Scope Clunat develops artificial intelligence, advanced AI systems, AGI-oriented research, and scientific technologies. This Governance Framework establishes the internal rules by which Clunat makes consequential decisions about research, models, products, data, security, deployment, partnerships, and organizational risk. It is intended to function as an operating system for responsible decision-making, not as a statement of aspirations alone. The framework applies across the company, including research, engineering, product, security, data, infrastructure, operations, commercial activity, and leadership. It covers the full AI lifecycle: problem selection, research, data acquisition, training, evaluation, red-teaming, deployment, monitoring, incident response, modification, retirement, and post-deployment learning. Clunat should treat governance as a technical and organizational capability. NIST's AI Risk Management Framework describes governance as a cross-cutting function that should be integrated throughout the AI lifecycle, while the OECD AI Principles emphasize accountability, transparency, robustness, security, safety, and systematic risk management. This framework adapts those ideas to an AI/AGI research company and adds stronger internal controls for advanced-model development. ### 2. Governance Principles Clunat adopts eight governing principles: - First, mission before momentum: Commercial pressure, benchmark competition, fundraising objectives, or publication incentives must not override a material safety, security, legal, or integrity concern. - Second, accountable ownership: Every consequential system, experiment, dataset, model release, and external commitment must have an identifiable owner with authority to act and a documented decision trail. - Third, independent challenge: Teams responsible for building a system should not be the only people deciding whether that system is safe or ready for release. Where risk is material, an independent evaluation function must have authority to block or delay deployment. - Fourth, proportionality: Governance should become stricter as capability, autonomy, scale, access, and potential harm increase. Low-risk experimentation should not carry frontier-model bureaucracy, but frontier systems should never be governed like ordinary software. - Fifth, evidence over intuition: Release and risk decisions should rely on evaluations, reproducible measurements, incident evidence, threat models, and documented uncertainty. - Sixth, human oversight and reversibility: Systems should have mechanisms for intervention, access restriction, rollback, shutdown, and safe decommissioning appropriate to their risk. - Seventh, scientific integrity: Clunat should preserve accurate records, disclose meaningful limitations, distinguish established findings from hypotheses, and avoid manipulating evaluations or selectively reporting results. - Eighth, continuous improvement: Governance is not a one-time approval. Policies, thresholds, evaluations, and organizational controls must evolve with evidence, capability changes, and external developments. ### 3. Governance Architecture The governance architecture consists of five layers: - The Board or Equivalent Governing Body: Provides fiduciary oversight, approves the company's risk appetite, oversees the most consequential AI risks, and receives periodic reporting on safety, security, legal exposure, major incidents, and strategic research risks. - The Executive Leadership Team: Converts board-level policy into operating priorities and is accountable for implementation. The Chief Executive or equivalent executive retains overall responsibility for organizational execution, but may not unilaterally override a formally established safety stop without documenting the rationale and obtaining the required governance review. - The AI Governance and Risk Committee: The central cross-functional decision body for high-impact AI matters. It includes leadership from research, engineering, security, legal/compliance, product, and safety/evaluation. Its responsibilities include model release gates, risk acceptance, high-severity incidents, exceptional research approvals, and material changes to safety controls. - The Independent Evaluation and Safety Function: Performs adversarial testing, capability evaluations, misuse testing, robustness analysis, and release-readiness reviews. Its independence is protected from product launch incentives. It maintains a direct escalation path to the AI Governance and Risk Committee and, for critical issues, to the board. - Research and Product Teams: Remain responsible for implementation and first-line risk management. Governance is therefore distributed: builders own day-to-day controls, independent evaluators challenge assumptions, and senior governance bodies make decisions on material residual risk. ### 4. Decision Rights and the Three-Lines Model Clunat operates a three-lines model: Line 1: Build and Operate: Research, engineering, product, infrastructure, and operations identify and manage risks in normal work. Each project maintains an owner, system description, risk register, evaluation plan, and change history. Line 2: Risk, Safety, Security, and Compliance: Independent functions define standards, review risk assessments, conduct challenge, monitor compliance, and escalate unresolved concerns. They have authority to require additional testing or controls. Line 3: Assurance: Periodic internal review or independent external assurance examines whether governance processes actually work. Assurance tests records, approvals, evaluation integrity, incident handling, access controls, and exceptions rather than simply checking whether policies exist. Decision rights must be explicit. A research lead may approve routine experiments within predefined boundaries. A release owner may approve low-risk releases that satisfy standard gates. High-impact or frontier-capability releases require committee approval. A critical safety or security issue can trigger an immediate deployment freeze without waiting for commercial approval. ### 5. AI Research Governance Research freedom is essential to Clunat, but research involving advanced capabilities requires structured controls. Every major research program should have a research charter specifying its objective, expected capabilities, known risks, datasets or tools involved, compute requirements, responsible owner, and intended publication or deployment path. Research should be classified by capability and risk. Categories include routine research, elevated-capability research, advanced capability research, and frontier/AGI-relevant research. The higher categories trigger stronger access controls, independent review, restricted environments, additional logging, and pre-defined stop conditions. Particular attention is given to research that could materially increase autonomous operation, cyber capability, biological or chemical assistance, large-scale persuasion, self-replication, automated exploitation, strategic deception, model autonomy, or the ability of systems to circumvent human oversight. The purpose is not to prohibit difficult science; it is to ensure that capability development and risk understanding progress together. Research records preserve hypotheses, experiment configurations, model versions, evaluation scripts, relevant failures, and material negative results. Reproducibility and traceability are governance assets because they allow the company to determine what changed and why a system behaves differently. ### 6. Model Development and Release Gates Every significant model passes a staged release process: Gate 0: Project Authorization: Define scope, owner, threat model, data provenance expectations, and evaluation plan. Gate 1: Development Readiness: Establish secure infrastructure, logging, access controls, and baseline evaluations. Gate 2: Capability Characterization: Measure intended and unintended capabilities, including meaningful uncertainty. Gate 3: Adversarial Evaluation: Conduct red-teaming, misuse testing, robustness testing, security testing, and targeted evaluations for the model's risk profile. Gate 4: Deployment Approval: Document residual risk, mitigations, access model, monitoring, rollback plan, and accountable approvers. Gate 5: Post-Deployment Review: Monitor incidents, drift, misuse, and unexpected capabilities to determine whether controls remain adequate. No model should be released merely because it performs well on benchmarks. Benchmark performance is evidence about capability, not proof of safety. Evaluation includes failure modes, distribution shift, adversarial prompting, tool use, autonomy, and system-level behavior where applicable. For advanced models, Clunat maintains explicit release thresholds. A threshold can be capability-based, risk-based, or both. If a model crosses a predefined capability boundary, the release process automatically escalates to enhanced governance even if the model is commercially attractive. ### 7. AGI and Frontier-System Governance AGI-oriented research requires a higher governance tier because the uncertainty around generality, autonomy, emergent capabilities, and downstream effects can be greater than for conventional AI products. Clunat establishes a Frontier Systems Protocol. Under this protocol, a system may be designated frontier when its capabilities, autonomy, scale, tool access, or potential impact exceed internally defined thresholds. Frontier designation triggers enhanced compute and infrastructure security, restricted access, independent evaluations, senior governance review, documented deployment assumptions, and continuous monitoring. Clunat avoids a single-number definition of AGI. Instead, governance evaluates a capability vector: breadth of tasks, reasoning performance, autonomy, persistence, tool use, ability to plan over long horizons, ability to modify or generate software, interaction with external systems, and evidence of novel or unexpected capabilities. For frontier systems, the company maintains a capability and risk dossier. It includes evaluation methodology, limitations, known failure modes, dangerous capability testing where appropriate, mitigations, residual risks, access restrictions, incident procedures, and release rationale. Where evaluations are incomplete, the uncertainty itself is recorded as a risk factor. ### 8. Safety, Security, and Misuse Prevention Safety and security are separate but interacting disciplines. AI safety addresses harmful behavior and system-level risk; security protects the model, infrastructure, weights, credentials, datasets, evaluation systems, and deployment environment from unauthorized access or manipulation. Clunat maintains threat models for major systems and updates them when capabilities or deployment conditions change. Critical assets use least-privilege access, strong authentication, secrets management, network segmentation, audit logging, secure software supply chains, and controlled model-weight access. Misuse controls are proportionate to the system. Depending on risk, controls include rate limits, user verification, content and tool-use policies, sandboxing, permission boundaries, human review, abuse monitoring, and restricted APIs. These controls are tested rather than assumed to work. Safety evaluation includes both ordinary and adversarial conditions. Red-team findings are tracked to closure, risk acceptance, or a documented decision not to remediate. Severe findings automatically trigger escalation and, where appropriate, suspension of deployment. ### 9. Data, Privacy, and Intellectual Property Governance Clunat maintains a documented data governance lifecycle covering acquisition, licensing, provenance, classification, storage, transformation, use, retention, and deletion. Training and evaluation datasets are traceable to their source and permitted use. Data that is not appropriate for a stated purpose does not enter the training pipeline merely because it is technically accessible. Metadata is separated from training content where practical, and datasets have machine-readable manifests describing origin, license or usage basis, preprocessing, exclusions, known limitations, and quality statistics. Data quality controls detect duplication, contamination, malicious content, personally identifiable information where relevant, and other unsuitable material. Privacy governance reflects applicable law and contractual commitments. Sensitive data receives stronger access controls and retention limits. Where personal data is processed, Clunat documents the purpose, lawful basis where applicable, safeguards, and deletion or correction mechanisms required by the applicable regime. Intellectual property is governed through documented ownership, licensing, contribution records, and review of external materials. Researchers must not knowingly misrepresent third-party material as original Clunat work. Open-source and research publication practices remain compatible with the security and licensing posture of the system. ### 10. Incident Response and Escalation Clunat maintains a unified AI incident process covering safety incidents, security breaches, privacy events, harmful misuse, evaluation failures, material model regressions, and governance violations. Incidents are classified by severity. Critical incidents include credible risks of severe harm, major unauthorized access to sensitive model assets, dangerous capability exposure, significant loss of control, or failures that materially undermine a high-impact deployment. Critical incidents require immediate containment, executive notification, preservation of evidence, and governance review. The incident process follows a defined sequence: detect, contain, assess, escalate, remediate, communicate, and learn. Containment can include disabling an endpoint, revoking credentials, restricting a model, freezing a release, or isolating infrastructure. Post-incident reviews identify root causes rather than assigning blame alone. Corrective actions have assigned owners and deadlines. Material incidents feed back into evaluation suites, threat models, engineering controls, and governance thresholds. ### 11. Ethics, Conflicts, and Whistleblowing Clunat maintains a professional ethics standard covering honesty in research, conflicts of interest, misuse of company resources, confidentiality, harassment, retaliation, manipulation of evaluations, and improper influence over governance decisions. Conflicts of interest must be disclosed when personal, financial, academic, or external relationships could reasonably affect a decision. The appropriate response may include recusal, independent review, or additional approval. Employees and collaborators have a secure channel to report concerns. Whistleblowers are protected from retaliation when raising concerns in good faith. A governance system that technically contains escalation channels but discourages their use is not functional. Evaluation independence is strictly enforced. Researchers must not be pressured to change test definitions, remove inconvenient failures, or delay reporting solely to make a model appear safer or more capable than the evidence supports. ### 12. Transparency, Documentation, and Auditability Governance decisions must be reconstructable after the fact. For material decisions, Clunat retains the decision, date, responsible individuals, evidence considered, alternatives, residual risks, conditions imposed, and follow-up requirements. Model cards or equivalent system documentation communicate intended use, limitations, evaluation results, major risks, access restrictions, and known failure modes. Internal dossiers contain more granular information than public documentation. Transparency does not require publishing dangerous operational details. Clunat distinguishes between public transparency, stakeholder transparency, and restricted internal records. Sensitive security information may require controlled disclosure, but the existence of a decision and its accountability remain recorded. Periodic governance audits examine whether release gates are being followed, whether exceptions are accumulating, whether incidents are closed effectively, and whether risk thresholds remain aligned with actual capabilities. ### 13. Risk Appetite and Exception Management Clunat maintains a written risk appetite approved by senior governance. The company accepts ordinary engineering uncertainty, research uncertainty, and bounded commercial risk; it does not casually accept unknown or unbounded risk involving severe potential harm, loss of control, major security compromise, or deliberate falsification of evidence. Exceptions must be explicit, time-bounded, owned, and approved at the appropriate governance level. An exception is not a way to bypass governance permanently. It is a controlled departure from a requirement with a documented rationale and compensating controls. Repeated exceptions trigger policy review. If a team continuously requires an exception, the underlying process may be unrealistic, or the organization may be systematically underestimating risk. ### 14. Scaling Governance with the Company Governance scales in maturity as Clunat grows. At the earliest stage, the founder and leadership group combine roles, but core controls: ownership, model inventory, risk classification, evaluation records, incident response, and release gates, exist from the beginning. As the company grows, Clunat separates research leadership from independent safety/evaluation leadership, formalizes board-level risk oversight, introduces internal audit or external assurance, and establishes specialized committees for security, privacy, research integrity, and frontier AI when justified. Governance is reviewed at least annually and after major capability changes, significant incidents, new regulatory obligations, material acquisitions, or major shifts in operational scope. The framework operates as a living control system rather than a static document. ### 15. Minimum Governance Controls Clunat maintains, at minimum: a current AI system inventory; named owners for significant systems; risk classifications; model and dataset provenance records; documented evaluation plans; release approvals; security access controls; incident response procedures; a whistleblowing channel; conflict-of-interest disclosures; exception registers; material decision logs; and periodic governance reviews. For advanced or frontier systems, minimum controls additionally include independent evaluation, explicit capability thresholds, enhanced infrastructure security, restricted access, red-team testing, documented residual risk, rollback or shutdown mechanisms, and senior-level release authorization. These controls are operational. The objective is to make the safe path the normal path: researchers know what is required, reviewers know what evidence to expect, and leadership knows when a decision must be escalated. ### 16. Governance Charter and Approval This Framework is adopted as the baseline governance standard for Clunat AI, AGI-oriented research, and related products. It is accompanied by detailed procedures, technical standards, model release checklists, incident playbooks, data governance standards, security policies, and research integrity procedures. The framework operates as an internal governance standard. Applicable law, regulation, contractual obligations, and corporate governance documents take precedence where they impose stricter requirements. The central operating principle is direct: Clunat builds intelligence with the same discipline with which it builds infrastructure. Capability creates opportunity, but governance determines whether that capability can be developed and deployed responsibly. For an AI/AGI research company, governance is part of the research and engineering system itself. -------------------------------------------------------------------------------- ## Safety policy Slug: /governance-safety-policy Summary: Operational requirements, threat modeling, and incident response procedures. Last Updated: 2026 ### Frontier AI Safety & Security Policy *Clear protocols for defensive isolation, containment, and ethical red lines* Our safety policy defines rigorous operational limits for agentic tools. Clunat systems are strictly bound by sandbox isolation, cryptographic token gating, and user confirmation loops before executing high-impact modifications. We maintain a zero-tolerance policy regarding weaponization, mass surveillance enablement, self-replicating exploits, and unauthorized data exfiltration. - Runtime isolation for all code execution engines - Continuous anomalous behavior telemetry and automated kill switches - Responsible vulnerability disclosure policy with guaranteed researcher bounties -------------------------------------------------------------------------------- ## Terms of Use and Terms & Conditions Slug: /legal-terms-of-service Summary: Clunat AI, AGI & Research Platforms Last Updated: 2026 ### Important Notice These Terms of Use and Terms & Conditions ("Terms") constitute a legally binding agreement between you ("you", "your", or "User") and Clunat ("Clunat", "Company", "we", "us", or "our"). These Terms govern your access to and use of Clunat websites, applications, artificial intelligence systems, chat interfaces, APIs, developer platforms, research tools, software, datasets where expressly provided, agentic systems, models, documentation, content, and other products and services offered by Clunat from time to time (collectively, the "Services"). By accessing, registering for, purchasing, downloading, integrating, or otherwise using any Service, you acknowledge that you have read, understood, and agreed to these Terms. If you do not agree to these Terms, you must not access or use the Services. ### 1. Definitions - "Account": An account created to access one or more Clunat Services. - "API": An application programming interface provided by Clunat that permits a User or authorized software application to access an Clunat model, service, dataset, computation service, or other functionality. - "Clunat Content": Software, models, model weights, interfaces, documentation, trademarks, logos, text, graphics, system prompts, safety systems, evaluation systems, infrastructure, and other materials owned, controlled, or licensed by Clunat. - "Customer Content": Prompts, inputs, files, data, instructions, code, documents, media, configurations, and other materials submitted to a Service by or on behalf of a User. - "Output": Text, code, images, audio, structured data, analysis, suggestions, classifications, actions, or other results generated by an Clunat AI system in response to Customer Content or system instructions. - "Products": Individual commercial, research, developer, API, software, hardware, model, or AI services made available by Clunat. - "Services": The Products together with associated websites, applications, APIs, infrastructure, documentation, support, and related functionality. - "User": An individual, organization, developer, business, research institution, or other legal person using a Service. - "Subscription": A recurring or fixed-term paid plan for a Service. - "Usage Limits": Rate limits, token limits, storage limits, compute limits, request limits, account restrictions, or other usage constraints imposed by a Service. ### 2. Acceptance of These Terms Your use of any Service constitutes acceptance of these Terms. Where you are accepting these Terms on behalf of an organization, you represent that you have authority to bind that organization, that the organization accepts these Terms, and that your use complies with applicable laws. ### 3. Changes to the Terms Clunat may modify these Terms to reflect changes in Services, technology, law, security, or governance practices. Revised Terms become effective upon publication, and continued use represents acceptance. ### 4. Clunat Services Services include AI reasoning models, generative tools, developer APIs, inference services, agentic workflows, scientific tools, and research environments. Certain features may be experimental, preview, or research-only. ### 5. Artificial Intelligence Disclaimer and Professional Advice AI systems can produce inaccurate, incomplete, or misleading outputs. Users are responsible for independently validating outputs. Clunat outputs do not constitute legal, medical, financial, or regulated professional advice. ### 6. AI Autonomy and Agentic Systems Where systems perform multi-step actions or interact with external tools, users remain responsible for configuring appropriate permission boundaries, access constraints, and approval controls. ### 7. Account Registration and API Access Users must provide accurate registration details, safeguard credentials, and adhere to documented rate limits. Reselling unrestricted API access or attempting to bypass security constraints is prohibited. ### 8. Acceptable Use and Security Restrictions Services must not be used for malicious activities, malware distribution, unauthorized intrusion, fraud, or violation of rights. Reverse engineering model weights, proprietary architectures, or safety controls is strictly prohibited. ### 9. Customer Content, Data Rights, and Model Training Users retain ownership of their Customer Content. Processing of Customer Content is conducted in accordance with product configurations and applicable privacy policies. Enterprise agreements may provide dedicated data-exclusion guarantees. ### 10. Intellectual Property and Output Ownership Clunat does not claim ownership of generated Outputs. All Clunat Content, underlying models, weights, and infrastructure remain the exclusive property of Clunat and its licensors. ### 11. Frontier AI and Safety Circumvention Frontier models may require enhanced verification and research authorizations. Bypassing safety alignment filters, exploiting system mechanisms, or defeating access controls is prohibited. ### 12. Payments, Subscriptions, and Suspension Subscription pricing, usage fees, and billing periods are established at point of purchase. Clunat reserves the right to suspend or terminate accounts that present security, safety, or legal risks. ### 13. Warranties, Disclaimers, and Limitation of Liability Services are provided on an "as is" and "as available" basis. To the maximum extent permitted by law, Clunat disclaims all warranties and is not liable for indirect, incidental, or consequential damages. ### 14. Dispute Resolution and Governing Law Disputes shall first be addressed through good-faith resolution. These Terms are governed by applicable law, subject to mandatory statutory consumer protections. ### Product Schedules and Checklists Schedule 1: Product-Specific Terms Governs consumer interfaces, developer APIs, research tiers, frontier model access, and enterprise configurations. Schedule 2: Acceptable Use Summary Prohibits system attacks, fraud, unauthorized harm, data extraction, and security circumvention. Schedule 3: User Responsibility Checklist Requires human review of AI-generated code, verification of factual outputs, and risk-appropriate oversight. -------------------------------------------------------------------------------- ## Terms of Use Slug: /legal-terms-of-use Summary: Platform and Service Usage Guidelines Last Updated: 2026 ### 1. Platform and Model Access These Terms of Use govern access to and usage of Clunat research models, APIs, software evaluation sandboxes, developer tools, and technical interfaces. ### 2. Acceptable and Responsible Usage Users must comply with all established safety guidelines. Attempting to circumvent safety alignment filters, conduct unauthorized extraction of model weights, deploy automated scrapers, or utilize models for unlawful purposes is strictly prohibited. ### 3. AI Autonomy and Tool Use When deploying agentic systems or models equipped with external tool execution capabilities, users remain responsible for establishing appropriate boundaries, authentication permissions, and human oversight. ### 4. Service Modifications and Resource Management Clunat reserves the right to modify, rate-limit, or terminate access to experimental interfaces or research endpoints with reasonable notice to maintain system security and operational integrity. -------------------------------------------------------------------------------- ## Privacy Policy Slug: /legal-privacy-notice Summary: Clunat AI, Research & Infrastructure Last Updated: 2026 ### 1. Introduction Clunat is an artificial intelligence and research organization developing artificial intelligence systems, advanced AI models, AI infrastructure, developer technologies, research platforms, agentic systems, scientific tools, and future technologies associated with advanced and general artificial intelligence. This Privacy Policy explains how Clunat collects, receives, uses, processes, stores, transfers, protects, and otherwise handles information relating to individuals and organizations that interact with Clunat. This Privacy Policy applies to the official Clunat websites, applications, AI assistants, chat products, developer platforms, APIs, research environments, model services, software, documentation portals, account systems, enterprise services, communications, events, support systems, and other products or services that expressly incorporate this Privacy Policy. Together, these are referred to as the "Services." Our objective is to process personal information in a manner that is lawful, proportionate, secure, transparent, and consistent with the purpose for which it was collected. Where information is not necessary for a stated purpose, Clunat seeks to minimize its collection, retention, and use. ### 2. Responsibility for Information Clunat operates under strict data stewardship standards. Where required by law, Clunat designates a privacy lead, Data Protection Officer, or grievance officer to oversee compliance with applicable regulatory frameworks. ### 3. Information We Collect and Receive Clunat may collect account identifiers (name, email address, authentication credentials, organizational affiliation), user-provided prompts and customer content submitted to models, API request metadata, diagnostic telemetry, billing records, and support correspondence. Users decide much of the content submitted to AI systems. Users should avoid submitting sensitive, confidential, or proprietary information unless a specific enterprise product has been contracted and configured for that category of data. ### 4. Processing AI Inputs and Operational Use When you submit information to an Clunat system, that data is processed to interpret your request, generate outputs, execute workflows, provide API responses, and maintain system security. Information is also used to diagnose system faults, detect abusive activity, prevent fraud, and comply with legal obligations. ### 5. AI Model Training and Improvement Not all Clunat products use customer data for model training. Enterprise and API services operate under strict contractual configurations where customer content is excluded from training pipelines. Where research or consumer products permit improvement use, data is processed with privacy-preserving controls such as filtering, access restriction, de-identification, and retention limits. ### 6. Conversations, API Data, and Uploaded Files Conversation histories and uploaded files are stored in accordance with product settings and user deletion controls. API metadata is retained for billing, security, and capacity management. When content is deleted from active systems, residual copies in encrypted disaster-recovery backups are securely overwritten over time. ### 7. Technical Telemetry, Cookies, and Analytics Clunat automatically collects technical diagnostic telemetry (IP addresses, device parameters, crash reports, latency metrics) to secure and maintain infrastructure. Cookies and local storage technologies are utilized for authentication, security verification, and performance analysis. ### 8. Safety, Abuse Prevention, and Legal Compliance Information is processed to detect account compromise, API exploitation, jailbreak attempts, and security threats. Clunat cooperates with lawful governmental processes while implementing legal reviews to protect user privacy against overbroad disclosures. ### 9. Information Sharing and International Transfers Clunat does not sell personal information as a commercial commodity. Data is shared only with vetted infrastructure, payment, and security subprocessors under binding contractual safeguards. Cross-border transfers adhere to applicable statutory mechanisms, including the DPDP framework and GDPR safeguards where relevant. ### 10. Data Security, Access Controls, and Retention We implement industry-standard encryption, strict least-privilege employee access controls, network segmentation, and regular vulnerability audits. Information is retained only as long as necessary to fulfill operational, research, accounting, security, and legal obligations. ### 11. Children's Privacy and Automated Decisions Services are not directed toward children below the statutory age permitted by applicable law. Clunat systems do not make solely automated, legally consequential decisions about individuals without human review and oversight. ### 12. Individual Privacy Rights and Grievance Redressal Depending on applicable jurisdiction, individuals possess rights to access, rectify, erase, or export their personal information, as well as rights to grievance redressal. Verified requests are processed within statutory timeframes. ### 13. Advanced AI Governance and Agentic Privacy For agentic systems and advanced model pipelines, privacy is integrated directly into tool execution boundaries and dataset provenance records. Autonomous agents operate with minimum required permissions, and memory mechanisms are subject to transparent user controls. -------------------------------------------------------------------------------- ## X Slug: /connect-x Summary: Follow our official updates, research papers, releases, and discussions on X. Last Updated: 2026 ### Official X Channel *Real-time announcements, technical threads, and team dispatches* Join our global developer and research community on X for the latest releases, changelogs, live demos, and open discussions around Clunat AI. -------------------------------------------------------------------------------- ## LinkedIn Slug: /connect-linkedin Summary: Connect with our organization, team members, open career positions, and industry updates. Last Updated: 2026 ### Clunat on LinkedIn *Career opportunities, technical executive briefings, and institutional research* Follow Clunat on LinkedIn to stay informed about career openings across research, engineering, and product, as well as institutional enterprise partnerships. -------------------------------------------------------------------------------- ## GitHub Slug: /connect-github Summary: Explore our open-source tools, SDKs, benchmarks, and community documentation. Last Updated: 2026 ### Open Source Repositories *Code, libraries, CLI utilities, and synthetic dataset loaders* We believe in advancing open science where safe and effective. Inspect our open-source SDKs, terminal runners, evaluation scripts, and community tools. - clunat-cli: Lightweight command-line engine for developer terminals - clunat-sdk-ts: Official TypeScript & Node.js client library - aligned-evals: Open evaluation test-suites for alignment verification -------------------------------------------------------------------------------- ## Hugging Face Slug: /connect-huggingface Summary: Weights, model cards, demo spaces, and public research checkpoints on Hugging Face. Last Updated: 2026 ### Organization on Hugging Face *Inspect model cards, evaluation metrics, and interactive demo spaces* Explore our public model checkpoints, tokenizer configurations, quantized weights, and spaces hosted directly on Hugging Face. - Model Checkpoints: Optimized 4-bit and 8-bit quantized weights - Safety-Datasets: Public alignment and safety preference sets - Interactive Spaces: Rapid prototyping and token latency evaluations -------------------------------------------------------------------------------- ## Clunat Bot Slug: /clunat-bot Summary: Autonomous verification crawler, web research agent, and epistemic knowledge indexer. Last Updated: March 2026 ### Overview & Directives *Clunat Bot operates as a sovereign research agent and epistemic verification crawler* Clunat Bot (identifier: ClunatBot/1.0) is an automated web agent and research crawler engineered by Clunat Research. It discovers, validates, and indexes scientific publications, open datasets, and formal verification proofs across the decentralized academic and developer web. Unlike traditional commercial web scrapers, Clunat Bot does not collect personal data, perform ad-targeting, or train models on unauthorized private content. Its purpose is strictly confined to epistemic citation verification, factual cross-validation of published research, and open dataset indexing under verified provenance guarantees. ### Crawler Specifications & User-Agent *Technical identification and request headers* Webmasters and system administrators can verify Clunat Bot requests via standard HTTP headers and cryptographic signatures: #### User-Agent String [HTTP Header] Mozilla/5.0 (compatible; ClunatBot/1.0; +https://clunat.com/clunat-bot) #### Politeness Rate Limits [Governance] Maximum 1–3 requests per second per domain; respects Crawl-delay and exponential backoff on HTTP 429/503. #### Robots.txt Adherence [Compliance] Strictly honors robots.txt, noindex tags, rel=nofollow, and LLM exclusion standards across all endpoints. #### Ephemeral Processing [Privacy] Zero-retention cache policy. Content is processed in memory for factual verification and discarded immediately. ### Verification & IP Whitelisting *How to authenticate genuine Clunat Bot requests* To verify that incoming requests originate from genuine Clunat Research clusters, execute reverse DNS lookups on the client IP address. Authentic requests resolve to *.crawler.clunat.com, with matching forward DNS validation. For webmasters with custom access controls or questions regarding crawl frequency, contact the research infrastructure desk at bot-ops@clunat.com. - Reverse DNS lookup pattern: *.crawler.clunat.com - Forward DNS verification: A record match confirmed - Dedicated contact channel: bot-ops@clunat.com -------------------------------------------------------------------------------- ## Clunat Bot Slug: /bot Summary: Autonomous verification crawler, web research agent, and epistemic knowledge indexer. Last Updated: March 2026 ### Overview & Directives *Clunat Bot operates as a sovereign research agent and epistemic verification crawler* Clunat Bot (identifier: ClunatBot/1.0) is an automated web agent and research crawler engineered by Clunat Research. It discovers, validates, and indexes scientific publications, open datasets, and formal verification proofs across the decentralized academic and developer web. Unlike traditional commercial web scrapers, Clunat Bot does not collect personal data, perform ad-targeting, or train models on unauthorized private content. Its purpose is strictly confined to epistemic citation verification, factual cross-validation of published research, and open dataset indexing under verified provenance guarantees. ### Crawler Specifications & User-Agent *Technical identification and request headers* Webmasters and system administrators can verify Clunat Bot requests via standard HTTP headers and cryptographic signatures: #### User-Agent String [HTTP Header] Mozilla/5.0 (compatible; ClunatBot/1.0; +https://clunat.com/clunat-bot) #### Politeness Rate Limits [Governance] Maximum 1–3 requests per second per domain; respects Crawl-delay and exponential backoff on HTTP 429/503. #### Robots.txt Adherence [Compliance] Strictly honors robots.txt, noindex tags, rel=nofollow, and LLM exclusion standards across all endpoints. #### Ephemeral Processing [Privacy] Zero-retention cache policy. Content is processed in memory for factual verification and discarded immediately. ### Verification & IP Whitelisting *How to authenticate genuine Clunat Bot requests* To verify that incoming requests originate from genuine Clunat Research clusters, execute reverse DNS lookups on the client IP address. Authentic requests resolve to *.crawler.clunat.com, with matching forward DNS validation. For webmasters with custom access controls or questions regarding crawl frequency, contact the research infrastructure desk at bot-ops@clunat.com. - Reverse DNS lookup pattern: *.crawler.clunat.com - Forward DNS verification: A record match confirmed - Dedicated contact channel: bot-ops@clunat.com -------------------------------------------------------------------------------- ## Clunat Bot Slug: /clunatbot Summary: Autonomous verification crawler, web research agent, and epistemic knowledge indexer. Last Updated: March 2026 ### Overview & Directives *Clunat Bot operates as a sovereign research agent and epistemic verification crawler* Clunat Bot (identifier: ClunatBot/1.0) is an automated web agent and research crawler engineered by Clunat Research. It discovers, validates, and indexes scientific publications, open datasets, and formal verification proofs across the decentralized academic and developer web. Unlike traditional commercial web scrapers, Clunat Bot does not collect personal data, perform ad-targeting, or train models on unauthorized private content. Its purpose is strictly confined to epistemic citation verification, factual cross-validation of published research, and open dataset indexing under verified provenance guarantees. ### Crawler Specifications & User-Agent *Technical identification and request headers* Webmasters and system administrators can verify Clunat Bot requests via standard HTTP headers and cryptographic signatures: #### User-Agent String [HTTP Header] Mozilla/5.0 (compatible; ClunatBot/1.0; +https://clunat.com/clunat-bot) #### Politeness Rate Limits [Governance] Maximum 1–3 requests per second per domain; respects Crawl-delay and exponential backoff on HTTP 429/503. #### Robots.txt Adherence [Compliance] Strictly honors robots.txt, noindex tags, rel=nofollow, and LLM exclusion standards across all endpoints. #### Ephemeral Processing [Privacy] Zero-retention cache policy. Content is processed in memory for factual verification and discarded immediately. ### Verification & IP Whitelisting *How to authenticate genuine Clunat Bot requests* To verify that incoming requests originate from genuine Clunat Research clusters, execute reverse DNS lookups on the client IP address. Authentic requests resolve to *.crawler.clunat.com, with matching forward DNS validation. For webmasters with custom access controls or questions regarding crawl frequency, contact the research infrastructure desk at bot-ops@clunat.com. - Reverse DNS lookup pattern: *.crawler.clunat.com - Forward DNS verification: A record match confirmed - Dedicated contact channel: bot-ops@clunat.com -------------------------------------------------------------------------------- ## About Clunat Slug: /about Summary: Intelligence for Human Flourishing Last Updated: 2026 ### We believe artificial general intelligence should help humanity learn, create, innovate, and grow safely. Clunat is an independent artificial intelligence and fundamental research laboratory dedicated to advancing cognitive systems, foundational reasoning architectures, and beneficial machine intelligence. We explore the foundations of increasingly capable AI systems, from how machines represent knowledge and reason over complex problems to how intelligence can become more efficient, general, reliable, and controllable. Our work sits at the intersection of artificial intelligence, mathematics, computer science, scientific computing, cognitive architectures, and AI safety. We pursue fundamental research not simply to build larger systems, but to understand the principles that make intelligence possible and to translate those principles into systems that can create meaningful value for humanity. "The goal of intelligence is not intelligence for its own sake. The goal is to expand what humanity can understand, create, discover, and achieve." ### Our Mission Our mission is to develop increasingly capable intelligence while advancing the scientific foundations required to make that intelligence useful, reliable, efficient, and safe. We are interested in the difficult questions underlying advanced AI: - How can a machine reason rather than merely predict? - How can an AI system construct, revise, and verify its own solutions? - How can intelligence become more computationally efficient without sacrificing capability? - How can models acquire broader and more transferable forms of reasoning? - How can an AI system understand uncertainty and recognize the limits of its own knowledge? - How can advanced intelligence remain interpretable, controllable, and aligned with human objectives as its capabilities increase? ### The Central Question "How can we build intelligence that expands the frontier of human possibility without compromising human agency?" ### The Four Pillars of Clunat 1. Learn We want AI to become a powerful instrument for human understanding. Clunat researches systems capable of synthesizing knowledge across disciplines, reasoning through complex problems, identifying relationships between ideas, and assisting humans in navigating increasingly large bodies of information. A truly useful intelligence should not merely tell a person what is already known. It should help them understand why something is true, explore competing explanations, identify missing information, connect concepts across domains, and construct new hypotheses. Our objective is not to replace human curiosity: it is to multiply it. 2. Create Intelligence should expand the creative potential of people. Clunat develops generative and reasoning technologies intended to augment researchers, developers, engineers, designers, writers, scientists, and creators. We believe the most powerful creative systems will not simply generate content on command. They will participate in an iterative process of thinking, exploring, critiquing, refining, and building alongside humans as cognitive collaborators. 3. Innovate The most important advances in artificial intelligence may emerge from solving problems that are not visible from the surface of today's systems. Clunat invests in fundamental research into the underlying mechanisms of intelligence: reasoning architectures, efficient attention, neural compression, memory, adaptive computation, and machine-assisted scientific discovery. Building increasingly capable systems cannot depend indefinitely on simply increasing computational scale. Intelligence must also become fundamentally more efficient in how it represents information, allocates computation, retrieves knowledge, and learns from experience. 4. Grow Safely Capability without safety is not progress. The development of increasingly powerful AI systems must be accompanied by an equally serious effort to understand, evaluate, and control those systems. Safety is not a layer added after research is complete: it is part of the research itself. Our approach combines alignment research, mechanistic investigation, adversarial evaluation, capability characterization, security engineering, and rigorous institutional governance. ### Our Research Philosophy At Clunat, we approach advanced AI as both a scientific problem and a societal responsibility. Ambitious research and responsible development are compatible: - Scientific exploration should remain bold. - Engineering should remain rigorous. - Evaluation should remain skeptical. - Governance should remain accountable. - Safety should remain inseparable from capability. ### Fundamental Research First Our foundation is research. We investigate underlying mechanisms that contribute to more capable forms of machine intelligence: - Representation: How intelligent systems encode information, concepts, and abstractions. - Reasoning: How systems perform structured inference over interconnected steps. - Memory: How intelligence retains, retrieves, and updates information efficiently. - Computation: Dynamically allocating compute to steps requiring deeper reasoning. - Efficiency: Operating with significantly lower computational and energy requirements. - Autonomy: Executing complex goals while bounded by explicit human permissions. - Safety: Understanding and constraining advanced systems before capabilities exceed evaluation. ### Building for a Global Scientific Community Clunat is proudly based in South Asia, India, while its research ambition is global. Scientific progress does not belong to a single country, institution, or company. We want to contribute to a broader ecosystem of researchers, engineers, students, developers, universities, laboratories, and independent thinkers working toward a deeper understanding of intelligence. Geography should not determine who gets to participate in the future of AI. ### The Clunat Standard "Do not build merely because something can be built. Build because it expands meaningful human capability, and build it with enough scientific and engineering discipline to understand its consequences." Progress occurs when capability is accompanied by understanding. Progress occurs when systems become more useful without becoming unnecessarily dangerous, when scientific knowledge becomes accessible, and when humanity gains new tools to solve problems previously beyond reach. Building intelligence for discovery, creation, and human flourishing. -------------------------------------------------------------------------------- ## About Clunat Slug: /about-clunat Summary: Intelligence for Human Flourishing Last Updated: 2026 ### We believe artificial general intelligence should help humanity learn, create, innovate, and grow safely. Clunat is an independent artificial intelligence and fundamental research laboratory dedicated to advancing cognitive systems, foundational reasoning architectures, and beneficial machine intelligence. We explore the foundations of increasingly capable AI systems, from how machines represent knowledge and reason over complex problems to how intelligence can become more efficient, general, reliable, and controllable. Our work sits at the intersection of artificial intelligence, mathematics, computer science, scientific computing, cognitive architectures, and AI safety. We pursue fundamental research not simply to build larger systems, but to understand the principles that make intelligence possible and to translate those principles into systems that can create meaningful value for humanity. "The goal of intelligence is not intelligence for its own sake. The goal is to expand what humanity can understand, create, discover, and achieve." ### Our Mission Our mission is to develop increasingly capable intelligence while advancing the scientific foundations required to make that intelligence useful, reliable, efficient, and safe. We are interested in the difficult questions underlying advanced AI: - How can a machine reason rather than merely predict? - How can an AI system construct, revise, and verify its own solutions? - How can intelligence become more computationally efficient without sacrificing capability? - How can models acquire broader and more transferable forms of reasoning? - How can an AI system understand uncertainty and recognize the limits of its own knowledge? - How can advanced intelligence remain interpretable, controllable, and aligned with human objectives as its capabilities increase? ### The Central Question "How can we build intelligence that expands the frontier of human possibility without compromising human agency?" ### The Four Pillars of Clunat 1. Learn We want AI to become a powerful instrument for human understanding. Clunat researches systems capable of synthesizing knowledge across disciplines, reasoning through complex problems, identifying relationships between ideas, and assisting humans in navigating increasingly large bodies of information. A truly useful intelligence should not merely tell a person what is already known. It should help them understand why something is true, explore competing explanations, identify missing information, connect concepts across domains, and construct new hypotheses. Our objective is not to replace human curiosity: it is to multiply it. 2. Create Intelligence should expand the creative potential of people. Clunat develops generative and reasoning technologies intended to augment researchers, developers, engineers, designers, writers, scientists, and creators. We believe the most powerful creative systems will not simply generate content on command. They will participate in an iterative process of thinking, exploring, critiquing, refining, and building alongside humans as cognitive collaborators. 3. Innovate The most important advances in artificial intelligence may emerge from solving problems that are not visible from the surface of today's systems. Clunat invests in fundamental research into the underlying mechanisms of intelligence: reasoning architectures, efficient attention, neural compression, memory, adaptive computation, and machine-assisted scientific discovery. Building increasingly capable systems cannot depend indefinitely on simply increasing computational scale. Intelligence must also become fundamentally more efficient in how it represents information, allocates computation, retrieves knowledge, and learns from experience. 4. Grow Safely Capability without safety is not progress. The development of increasingly powerful AI systems must be accompanied by an equally serious effort to understand, evaluate, and control those systems. Safety is not a layer added after research is complete: it is part of the research itself. Our approach combines alignment research, mechanistic investigation, adversarial evaluation, capability characterization, security engineering, and rigorous institutional governance. ### Our Research Philosophy At Clunat, we approach advanced AI as both a scientific problem and a societal responsibility. Ambitious research and responsible development are compatible: - Scientific exploration should remain bold. - Engineering should remain rigorous. - Evaluation should remain skeptical. - Governance should remain accountable. - Safety should remain inseparable from capability. ### Fundamental Research First Our foundation is research. We investigate underlying mechanisms that contribute to more capable forms of machine intelligence: - Representation: How intelligent systems encode information, concepts, and abstractions. - Reasoning: How systems perform structured inference over interconnected steps. - Memory: How intelligence retains, retrieves, and updates information efficiently. - Computation: Dynamically allocating compute to steps requiring deeper reasoning. - Efficiency: Operating with significantly lower computational and energy requirements. - Autonomy: Executing complex goals while bounded by explicit human permissions. - Safety: Understanding and constraining advanced systems before capabilities exceed evaluation. ### Building for a Global Scientific Community Clunat is proudly based in South Asia, India, while its research ambition is global. Scientific progress does not belong to a single country, institution, or company. We want to contribute to a broader ecosystem of researchers, engineers, students, developers, universities, laboratories, and independent thinkers working toward a deeper understanding of intelligence. Geography should not determine who gets to participate in the future of AI. ### The Clunat Standard "Do not build merely because something can be built. Build because it expands meaningful human capability, and build it with enough scientific and engineering discipline to understand its consequences." Progress occurs when capability is accompanied by understanding. Progress occurs when systems become more useful without becoming unnecessarily dangerous, when scientific knowledge becomes accessible, and when humanity gains new tools to solve problems previously beyond reach. Building intelligence for discovery, creation, and human flourishing. -------------------------------------------------------------------------------- ## The Four Pillars of Clunat Slug: /pillars Summary: Learn, Create, Innovate, and Grow Safely. Last Updated: 2026 ### 1. Learn We want AI to become a powerful instrument for human understanding. Clunat researches systems capable of synthesizing knowledge across disciplines, reasoning through complex problems, identifying relationships between ideas, and assisting humans in navigating increasingly large bodies of information. A truly useful intelligence should not merely tell a person what is already known. It should help them understand why something is true, explore competing explanations, identify missing information, connect concepts across domains, and construct new hypotheses. Our objective is not to replace human curiosity: it is to multiply it. ### 2. Create Intelligence should expand the creative potential of people. Clunat develops generative and reasoning technologies intended to augment researchers, developers, engineers, designers, writers, scientists, and creators. We believe the most powerful creative systems will not simply generate content on command. They will participate in an iterative process of thinking, exploring, critiquing, refining, and building alongside humans as cognitive collaborators. ### 3. Innovate The most important advances in artificial intelligence may emerge from solving problems that are not visible from the surface of today's systems. Clunat invests in fundamental research into the underlying mechanisms of intelligence: reasoning architectures, efficient attention, neural compression, memory, adaptive computation, and machine-assisted scientific discovery. Building increasingly capable systems cannot depend indefinitely on simply increasing computational scale. Intelligence must also become fundamentally more efficient in how it represents information, allocates computation, retrieves knowledge, and learns from experience. ### 4. Grow Safely Capability without safety is not progress. The development of increasingly powerful AI systems must be accompanied by an equally serious effort to understand, evaluate, and control those systems. Safety is not a layer added after research is complete: it is part of the research itself. Our approach combines alignment research, mechanistic investigation, adversarial evaluation, capability characterization, security engineering, and rigorous institutional governance. -------------------------------------------------------------------------------- ## Our Research Philosophy Slug: /philosophy Summary: Approaching advanced AI as both a scientific problem and a societal responsibility. Last Updated: 2026 ### Research Principles At Clunat, we approach advanced AI as both a scientific problem and a societal responsibility. Ambitious research and responsible development are compatible: - Scientific exploration should remain bold. - Engineering should remain rigorous. - Evaluation should remain skeptical. - Governance should remain accountable. - Safety should remain inseparable from capability. ### Fundamental Research First Our foundation is research. We investigate underlying mechanisms that contribute to more capable forms of machine intelligence: - Representation: How intelligent systems encode information, concepts, and abstractions. - Reasoning: How systems perform structured inference over interconnected steps. - Memory: How intelligence retains, retrieves, and updates information efficiently. - Computation: Dynamically allocating compute to steps requiring deeper reasoning. - Efficiency: Operating with significantly lower computational and energy requirements. - Autonomy: Executing complex goals while bounded by explicit human permissions. - Safety: Understanding and constraining advanced systems before capabilities exceed evaluation. ### The Clunat Standard "Do not build merely because something can be built. Build because it expands meaningful human capability, and build it with enough scientific and engineering discipline to understand its consequences." Progress occurs when capability is accompanied by understanding. Progress occurs when systems become more useful without becoming unnecessarily dangerous, when scientific knowledge becomes accessible, and when humanity gains new tools to solve problems previously beyond reach. Building intelligence for discovery, creation, and human flourishing. -------------------------------------------------------------------------------- ## Index Slug: /index Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Index Slug: /research Summary: Scientific papers, technical reports, and foundational research publications. Last Updated: 2026 ### Research Index The Clunat research index documents foundational inquiries into cognitive systems, reasoning architectures, computational efficiency, and safe machine intelligence. Publications, technical reports, and preprints will be indexed here as they are released. -------------------------------------------------------------------------------- ## Partner Network Services Slug: /partners Summary: A global alliance of certified system integrators, technical consultancies, and independent software vendors deploying frontier Clunat cognitive architectures at enterprise scale. Last Updated: 2026 ### Accelerating Safe Enterprise AI Adoption *Bridging frontier foundation models with mission-critical corporate operations* The Clunat Partner Network (CPN) connects leading global system integrators, independent software vendors (ISVs), and domain-specialized consultancies with the advanced cognitive infrastructure of Clunat. As organizations transition from experimental pilots to production-grade agentic systems, our partner network provides the strategic advisory, custom model adaptation, sovereign data isolation, and aligned AI safety engineering necessary for dependable real-world execution. - Direct technical co-engineering and architectural reviews with Clunat foundation researchers - Priority access to unreleased model weights, high-concurrency inference endpoints, and specialized fine-tuning toolchains - Formal Aligned AI audit, verification, and red-teaming certification programs - Co-selling enablement, shared enterprise pipelines, and dedicated executive partner sponsorship ### Partnership Tracks *Tailored programs engineered for distinct technical and commercial collaboration models* We offer three specialized partnership tracks to support diverse organizational capabilities and client engagements: #### Global System Integrators & Consultancies [Advisory & Systems Integration] For enterprise advisory firms and digital transformation leaders delivering multi-agent workflows, legacy ERP integrations, and sovereign on-premises deployments. #### Independent Software Vendors (ISVs) [Product Integration] For SaaS providers and product teams embedding Clunat foundation reasoning, automated code generation, and multi-modal intelligence into commercial platforms. #### Applied Engineering & Research Boutiques [R&D Collaboration] For specialized AI engineering labs and academic centers co-developing domain-specific cognitive benchmarks, automated safety monitors, and specialized tool-use pipelines. ### Partner Enablement & Infrastructure *Tools, sandboxes, and engineering resources provided to verified partners* Verified partners operate within an accelerated technical environment designed to eliminate deployment friction and ensure predictable performance under stringent enterprise conditions. - Dedicated Developer Sandboxes: High-throughput, isolated environments for latency profiling, load testing, and edge deployment validation. - Direct Engineering Channels: Direct communication channels with Clunat core infrastructure leads and bi-weekly technical syncs. - Guaranteed Enterprise SLAs: Priority model routing, reserved GPU inference clusters, and guaranteed uptime commitments for mission-critical workloads. - Joint Go-To-Market Programs: Co-branded whitepapers, customer case studies, and presentation opportunities at global industry conferences. ### Quality, Safety & Certification Standards *Upholding aligned AI safety and ethical governance across all client deliverables* Every member of the Clunat Partner Network commits to our core Safety Policy and Responsible Scaling Framework. System architectures must satisfy rigorous audit procedures before receiving certification. - Zero-Retention Guarantee: Customer data submitted through partner implementations is never logged or utilized for base foundation model training. - Aligned AI Red-Teaming: Standardized automated safety harnesses to audit prompt boundaries, bias, and output veracity prior to production sign-off. - Annual Technical Recertification: Continuous credentialing to ensure partner engineering teams remain proficient in the latest Clunat model releases. ### Join the Partner Network *Submit your organization credentials to begin the accreditation review* Whether your organization is seeking to become a certified implementation partner or looking to engage an accredited Clunat integrator for your enterprise deployment, our partner alliances team is here to assist. Partner Alliances: partners@clunat.com | Developer Relations: dev@clunat.com | Official Portal: platform.clunat.com -------------------------------------------------------------------------------- ## Partner Network Services Slug: /partner-directory Summary: A global alliance of certified system integrators, technical consultancies, and independent software vendors deploying frontier Clunat cognitive architectures at enterprise scale. Last Updated: 2026 ### Accelerating Safe Enterprise AI Adoption *Bridging frontier foundation models with mission-critical corporate operations* The Clunat Partner Network (CPN) connects leading global system integrators, independent software vendors (ISVs), and domain-specialized consultancies with the advanced cognitive infrastructure of Clunat. As organizations transition from experimental pilots to production-grade agentic systems, our partner network provides the strategic advisory, custom model adaptation, sovereign data isolation, and aligned AI safety engineering necessary for dependable real-world execution. - Direct technical co-engineering and architectural reviews with Clunat foundation researchers - Priority access to unreleased model weights, high-concurrency inference endpoints, and specialized fine-tuning toolchains - Formal Aligned AI audit, verification, and red-teaming certification programs - Co-selling enablement, shared enterprise pipelines, and dedicated executive partner sponsorship ### Partnership Tracks *Tailored programs engineered for distinct technical and commercial collaboration models* We offer three specialized partnership tracks to support diverse organizational capabilities and client engagements: #### Global System Integrators & Consultancies [Advisory & Systems Integration] For enterprise advisory firms and digital transformation leaders delivering multi-agent workflows, legacy ERP integrations, and sovereign on-premises deployments. #### Independent Software Vendors (ISVs) [Product Integration] For SaaS providers and product teams embedding Clunat foundation reasoning, automated code generation, and multi-modal intelligence into commercial platforms. #### Applied Engineering & Research Boutiques [R&D Collaboration] For specialized AI engineering labs and academic centers co-developing domain-specific cognitive benchmarks, automated safety monitors, and specialized tool-use pipelines. ### Partner Enablement & Infrastructure *Tools, sandboxes, and engineering resources provided to verified partners* Verified partners operate within an accelerated technical environment designed to eliminate deployment friction and ensure predictable performance under stringent enterprise conditions. - Dedicated Developer Sandboxes: High-throughput, isolated environments for latency profiling, load testing, and edge deployment validation. - Direct Engineering Channels: Direct communication channels with Clunat core infrastructure leads and bi-weekly technical syncs. - Guaranteed Enterprise SLAs: Priority model routing, reserved GPU inference clusters, and guaranteed uptime commitments for mission-critical workloads. - Joint Go-To-Market Programs: Co-branded whitepapers, customer case studies, and presentation opportunities at global industry conferences. ### Quality, Safety & Certification Standards *Upholding aligned AI safety and ethical governance across all client deliverables* Every member of the Clunat Partner Network commits to our core Safety Policy and Responsible Scaling Framework. System architectures must satisfy rigorous audit procedures before receiving certification. - Zero-Retention Guarantee: Customer data submitted through partner implementations is never logged or utilized for base foundation model training. - Aligned AI Red-Teaming: Standardized automated safety harnesses to audit prompt boundaries, bias, and output veracity prior to production sign-off. - Annual Technical Recertification: Continuous credentialing to ensure partner engineering teams remain proficient in the latest Clunat model releases. ### Join the Partner Network *Submit your organization credentials to begin the accreditation review* Whether your organization is seeking to become a certified implementation partner or looking to engage an accredited Clunat integrator for your enterprise deployment, our partner alliances team is here to assist. Partner Alliances: partners@clunat.com | Developer Relations: dev@clunat.com | Official Portal: platform.clunat.com -------------------------------------------------------------------------------- ## Partner Network Services Slug: /ecosystem-partner-network Summary: A global alliance of certified system integrators, technical consultancies, and independent software vendors deploying frontier Clunat cognitive architectures at enterprise scale. Last Updated: 2026 ### Accelerating Safe Enterprise AI Adoption *Bridging frontier foundation models with mission-critical corporate operations* The Clunat Partner Network (CPN) connects leading global system integrators, independent software vendors (ISVs), and domain-specialized consultancies with the advanced cognitive infrastructure of Clunat. As organizations transition from experimental pilots to production-grade agentic systems, our partner network provides the strategic advisory, custom model adaptation, sovereign data isolation, and aligned AI safety engineering necessary for dependable real-world execution. - Direct technical co-engineering and architectural reviews with Clunat foundation researchers - Priority access to unreleased model weights, high-concurrency inference endpoints, and specialized fine-tuning toolchains - Formal Aligned AI audit, verification, and red-teaming certification programs - Co-selling enablement, shared enterprise pipelines, and dedicated executive partner sponsorship ### Partnership Tracks *Tailored programs engineered for distinct technical and commercial collaboration models* We offer three specialized partnership tracks to support diverse organizational capabilities and client engagements: #### Global System Integrators & Consultancies [Advisory & Systems Integration] For enterprise advisory firms and digital transformation leaders delivering multi-agent workflows, legacy ERP integrations, and sovereign on-premises deployments. #### Independent Software Vendors (ISVs) [Product Integration] For SaaS providers and product teams embedding Clunat foundation reasoning, automated code generation, and multi-modal intelligence into commercial platforms. #### Applied Engineering & Research Boutiques [R&D Collaboration] For specialized AI engineering labs and academic centers co-developing domain-specific cognitive benchmarks, automated safety monitors, and specialized tool-use pipelines. ### Partner Enablement & Infrastructure *Tools, sandboxes, and engineering resources provided to verified partners* Verified partners operate within an accelerated technical environment designed to eliminate deployment friction and ensure predictable performance under stringent enterprise conditions. - Dedicated Developer Sandboxes: High-throughput, isolated environments for latency profiling, load testing, and edge deployment validation. - Direct Engineering Channels: Direct communication channels with Clunat core infrastructure leads and bi-weekly technical syncs. - Guaranteed Enterprise SLAs: Priority model routing, reserved GPU inference clusters, and guaranteed uptime commitments for mission-critical workloads. - Joint Go-To-Market Programs: Co-branded whitepapers, customer case studies, and presentation opportunities at global industry conferences. ### Quality, Safety & Certification Standards *Upholding aligned AI safety and ethical governance across all client deliverables* Every member of the Clunat Partner Network commits to our core Safety Policy and Responsible Scaling Framework. System architectures must satisfy rigorous audit procedures before receiving certification. - Zero-Retention Guarantee: Customer data submitted through partner implementations is never logged or utilized for base foundation model training. - Aligned AI Red-Teaming: Standardized automated safety harnesses to audit prompt boundaries, bias, and output veracity prior to production sign-off. - Annual Technical Recertification: Continuous credentialing to ensure partner engineering teams remain proficient in the latest Clunat model releases. ### Join the Partner Network *Submit your organization credentials to begin the accreditation review* Whether your organization is seeking to become a certified implementation partner or looking to engage an accredited Clunat integrator for your enterprise deployment, our partner alliances team is here to assist. Partner Alliances: partners@clunat.com | Developer Relations: dev@clunat.com | Official Portal: platform.clunat.com -------------------------------------------------------------------------------- ## Partner Network Services Slug: /partner-services Summary: A global alliance of certified system integrators, technical consultancies, and independent software vendors deploying frontier Clunat cognitive architectures at enterprise scale. Last Updated: 2026 ### Accelerating Safe Enterprise AI Adoption *Bridging frontier foundation models with mission-critical corporate operations* The Clunat Partner Network (CPN) connects leading global system integrators, independent software vendors (ISVs), and domain-specialized consultancies with the advanced cognitive infrastructure of Clunat. As organizations transition from experimental pilots to production-grade agentic systems, our partner network provides the strategic advisory, custom model adaptation, sovereign data isolation, and aligned AI safety engineering necessary for dependable real-world execution. - Direct technical co-engineering and architectural reviews with Clunat foundation researchers - Priority access to unreleased model weights, high-concurrency inference endpoints, and specialized fine-tuning toolchains - Formal Aligned AI audit, verification, and red-teaming certification programs - Co-selling enablement, shared enterprise pipelines, and dedicated executive partner sponsorship ### Partnership Tracks *Tailored programs engineered for distinct technical and commercial collaboration models* We offer three specialized partnership tracks to support diverse organizational capabilities and client engagements: #### Global System Integrators & Consultancies [Advisory & Systems Integration] For enterprise advisory firms and digital transformation leaders delivering multi-agent workflows, legacy ERP integrations, and sovereign on-premises deployments. #### Independent Software Vendors (ISVs) [Product Integration] For SaaS providers and product teams embedding Clunat foundation reasoning, automated code generation, and multi-modal intelligence into commercial platforms. #### Applied Engineering & Research Boutiques [R&D Collaboration] For specialized AI engineering labs and academic centers co-developing domain-specific cognitive benchmarks, automated safety monitors, and specialized tool-use pipelines. ### Partner Enablement & Infrastructure *Tools, sandboxes, and engineering resources provided to verified partners* Verified partners operate within an accelerated technical environment designed to eliminate deployment friction and ensure predictable performance under stringent enterprise conditions. - Dedicated Developer Sandboxes: High-throughput, isolated environments for latency profiling, load testing, and edge deployment validation. - Direct Engineering Channels: Direct communication channels with Clunat core infrastructure leads and bi-weekly technical syncs. - Guaranteed Enterprise SLAs: Priority model routing, reserved GPU inference clusters, and guaranteed uptime commitments for mission-critical workloads. - Joint Go-To-Market Programs: Co-branded whitepapers, customer case studies, and presentation opportunities at global industry conferences. ### Quality, Safety & Certification Standards *Upholding aligned AI safety and ethical governance across all client deliverables* Every member of the Clunat Partner Network commits to our core Safety Policy and Responsible Scaling Framework. System architectures must satisfy rigorous audit procedures before receiving certification. - Zero-Retention Guarantee: Customer data submitted through partner implementations is never logged or utilized for base foundation model training. - Aligned AI Red-Teaming: Standardized automated safety harnesses to audit prompt boundaries, bias, and output veracity prior to production sign-off. - Annual Technical Recertification: Continuous credentialing to ensure partner engineering teams remain proficient in the latest Clunat model releases. ### Join the Partner Network *Submit your organization credentials to begin the accreditation review* Whether your organization is seeking to become a certified implementation partner or looking to engage an accredited Clunat integrator for your enterprise deployment, our partner alliances team is here to assist. Partner Alliances: partners@clunat.com | Developer Relations: dev@clunat.com | Official Portal: platform.clunat.com --------------------------------------------------------------------------------