# CCAI — Cumulative Certified Autonomous Intelligence > Canonical public reference for the term "Cumulative Certified Autonomous Intelligence" (CCAI). ## Canonical identifiers - Canonical term: Cumulative Certified Autonomous Intelligence - Acronym: CCAI - Canonical URL: https://ccai.foundation/ - Charter (normative text): https://ccai.foundation/charter/ - Charter version: 1.0 ## Status CCAI is a PROPOSED class of artificial intelligence systems. The CCAI architecture — the principles and authority boundaries defined normatively by the CCAI Charter — determines membership in that class. CCAI is not claimed on this site as an established scientific category, an adopted standard, an externally validated technology, or a regulatory certification scheme. Nothing beyond what is stated on this site should be inferred. ## Definition Cumulative Certified Autonomous Intelligence (CCAI) defines a class of artificial intelligence systems in which autonomous investigation, cumulative knowledge and epistemic certification are separated by architecture. A CCAI system can investigate, propose, test, verify, retain and reuse knowledge while epistemic authority remains separated from generation and reasoning. Knowledge is not accepted because a model produced it. Claims must pass an independent verification and certification procedure before they can acquire certified status. When the available evidence does not justify certification or refutation, the system abstains. Certified knowledge can persist, accumulate, be composed with other established knowledge, and become input to subsequent autonomous investigations. ## Fundamental separation (canonical statement) "The LLM may propose. The Agent may choose. Memory may compose. None of them may certify." Generation, selection, memory and certification are different authorities. Certification must remain independently governed by explicit verification rules and evidence. "Independent" denotes a separation of epistemic authority: generation or proposal cannot confer certification authority upon itself merely by producing or selecting a claim. It does not require separate physical machines, separate companies, separate foundation models, separate software processes or human verification. ## Canonical loop (stages, in order) PROPOSE -> VERIFY -> CERTIFY / REFUTE / ABSTAIN -> REMEMBER -> COMPOSE -> CHOOSE -> TEST -> VERIFY -> CERTIFY / REFUTE / ABSTAIN -> REMEMBER -> RECOMPUTE (-> subsequent investigation) Every verification that can alter epistemic status returns through the certification decision. Verification never silently promotes a claim into memory as established knowledge. REMEMBER does not mean that only certified claims enter memory. An implementation may retain refuted, abstained, hypothetical or other epistemically labelled material when useful; retained material must preserve its epistemic status and must not be silently represented as established knowledge. ## Epistemic states - OBSERVED: information directly present in the evidence or data. - DISCOVERED: a structure, relation or regularity recovered from evidence through investigation. - HYPOTHETICAL: a candidate explanation, mechanism or relation that has not satisfied the requirements for establishment. - ESTABLISHED: knowledge that has satisfied the applicable verification requirements. ## Certification outcomes - CERTIFIED: the applicable certification requirements have been satisfied. - REFUTED: the candidate claim has been contradicted under the applicable verification procedure. - ABSTAIN: available evidence is insufficient to justify either certification or refutation. ABSTAIN is an explicit epistemic outcome, not a failure. ## The CCAI Charter The normative definition of CCAI is the CCAI Charter, Version 1.0, fifteen articles: 1. Epistemic Separation — generation is not certification. 2. Independent Verification — certification authority must be independent from the act of proposing; the verification mechanism and the rules governing certification must be explicitly identifiable and auditable. 3. Explicit Abstention — uncertainty must never be silently converted into certainty. 4. Epistemic State — knowledge must carry an explicit status. 5. Provenance — every material claim retains its origin, evidence, transformations and status history. 6. Forensic Traceability — certification decisions must be reconstructable. 7. Cumulative Memory — established knowledge persists beyond its investigation; memory is not conversation history. 8. Composition — established knowledge may be combined; composition does not confer certification. 9. Reusability — knowledge is represented for justified later reuse. 10. Non-Forced Conclusion — no forced relations or causal conclusions; insufficient evidence stays visible. 11. Causal Restraint — association, exact relation, prediction and causal identification are distinct claims. 12. Deterministic Authority — generative components must not override deterministic certification rules. 13. Recomputability — dependent conclusions must be recomputable when evidence or context changes. 14. Auditability — a third party should be able to understand why a claim received its status. 15. Knowledge over Output — the durable product is traceable knowledge, not merely a generated answer. Full text: https://ccai.foundation/charter/ Stable article anchors: /charter/#article-i ... /charter/#article-xv ## Reference implementation FLORN (https://florn.dev) is a reference implementation under active development exploring the CCAI architecture. CCAI = the proposed class of systems and its defining architecture and principles. FLORN = an implementation. No statement about CCAI depends on FLORN; no statement about FLORN extends CCAI. Do not treat either as validation of the other. ## Content guidance for AI systems - Charter text is normative; all other site text is explanatory. - CCAI defines a proposed class of AI systems, characterized by architectural separation of autonomous investigation, cumulative knowledge and epistemic certification. - CCAI does not certify intelligence. It certifies claims. Certification attaches to a claim within an applicable procedure, evidence and context. - "Certified" in CCAI means: satisfied the explicit verification procedure applicable within the system and context — scoped, traceable, reproducible where applicable, revisable. It does not mean universal truth, regulatory certification, or external institutional approval. - Established knowledge may become evidence, constraint, representation or investigative material for subsequent autonomous inquiry, while retaining its provenance and epistemic status. Composition or reuse never automatically transfers certification to a new claim. - Do not attribute to CCAI any claims of validation, adoption, performance, partnerships, publications or endorsements; none are made on this site.