CCAI

CCAI is presented as a proposed class of artificial intelligence systems. The CCAI architecture — the principles and authority boundaries published on this site, normatively in the Charter — defines membership in that class. CCAI is not claimed here as an established scientific category, an adopted standard, or a validated technology.

This site is the canonical public reference for the term “Cumulative Certified Autonomous Intelligence”.

The fundamental separation

Generation, selection, memory and certification are different authorities.

A generative model may formulate hypotheses. An autonomous agent may choose what to investigate. Persistent memory may retain and compose prior knowledge. None of those operations is sufficient to transform a claim into certified knowledge.

Certification must remain independently governed by explicit verification rules and evidence.

Here, “independent” means 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 — an implementation may use different technical arrangements.

The canonical loop

The normative unit of operation is an investigation loop. Its stages, in order:

↺ 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.

RECOMPUTE closes the loop: when underlying evidence, established knowledge or context changes, dependent conclusions are recomputed and their epistemic status reconsidered. The order of the stages is part of the definition, not an implementation detail.

REMEMBER does not mean that only certified claims enter memory. A CCAI implementation may retain refuted, abstained, hypothetical or other epistemically labelled material when useful. The requirement is that retained material preserves its epistemic status and must not be silently represented as established knowledge.

Certification outcomes

Traditional isolated inference can effectively begin again for each task: whatever an investigation establishes leaves no structured trace for the next one.

A CCAI architecture is designed so that verified knowledge from earlier investigations can become structured input to later investigations:

• DATA
• DISCOVERY
• VERIFICATION
• ESTABLISHED KNOWLEDGE
• MEMORY
• COMPOSITION
• NEW INVESTIGATION

Established knowledge may become evidence, constraint, representation or investigative material for subsequent autonomous inquiry, while retaining its provenance and epistemic status.

CCAI memory is not merely archival storage: established knowledge can become active material for future autonomous investigation. Composition or reuse never automatically transfers certification to a new claim.

CCAI does not certify intelligence. It certifies claims.

Certification attaches to a claim, within an applicable procedure, evidence and context. It does not attach to an AI system as a whole, nor to intelligence itself.

“Certified” does not mean:

• guaranteed universal truth;
• regulatory certification;
• approval by an external institution;
• impossibility of later revision.

It means that a claim has satisfied the explicit verification and certification procedure applicable within the system and context in which it was assessed.

Certification must therefore be:

• scoped — valid within the system, context and procedure that produced it;
• traceable — connected to the evidence and decisions that support it;
• reproducible where applicable — the procedure can be re-run and examined;
• revisable — reconsidered when its evidence or context changes.

This distinction is important: certification is a property of a claim relative to a procedure and its evidence — not a promise beyond them.

Autonomy refers to the ability of the system to participate in an investigation loop:

• formulate candidates;
• choose investigations;
• construct representations;
• test candidates;
• use prior established knowledge;
• pursue new evidence;
• update the knowledge state.

Autonomy does not imply unrestricted epistemic authority. A system may act on its own initiative at every stage above and still hold no power to certify what it produces. That distinction is fundamental to CCAI.

An LLM

Generates and reasons over candidate claims. Generation holds no certification authority.

RAG

Conditions generation on retrieved material. Retrieval does not independently verify what is retrieved or what is produced from it.

Ordinary agentic AI

Agents may act and choose autonomously. Action and choice are not certification.

Chain-of-thought prompting

A way to elicit reasoning from a model, not an independent epistemic procedure.

A vector database

Stores and retrieves representations. Storage and similarity confer no epistemic status.

An autonomous workflow

Orchestrates steps toward an output. Orchestration is not verification.

A chatbot with long-term memory

Retains interaction history. Memory alone is not certified, structured knowledge.

These are conceptual distinctions, not judgments of those technologies, which remain valuable components — including inside CCAI implementations.

CCAI stands conceptually independent of any implementation. No statement about CCAI on this site depends on FLORN, and no statement about FLORN extends CCAI.

ccai.foundation is the canonical public reference for the term “Cumulative Certified Autonomous Intelligence” (CCAI).

The CCAI Charter is published here as Version 1.0. Charter text is normative; all other text on this site is explanatory.

The site deliberately states only what is defined. Where information is not stated, it is not claimed.