AI Integration

AIEP provides a machine-readable evidence surface so that AI systems can retrieve, cite, validate, and train on AIEP-governed artefacts. This section covers the technical surface for AI-to-AIEP integration.

The machine-readable surface

AIEP exposes a structured, schema-validated, version-stable API surface that AI systems can query directly. All queries resolve to cryptographically commited artefacts — not stochastic responses.

The canonical endpoint:

/.well-known/aiep/
  schemas/aiep.canonical.schema.v3.0.0.json
  capabilities.json
  status.json

Every deployed AIEP substrate also exposes:

GET /aiep/v1/artefact/{sha256}          → canonical artefact record
GET /aiep/v1/evidence/{claim_id}        → full evidence chain
GET /aiep/v1/certificate/{cert_hash}    → certification record
GET /aiep/v1/swarm/status               → node health

Full specification: Machine Interface →

AI Retrieval — P60, P61, P62

AI Retrieval →

When an AI system queries an AIEP substrate, the response includes:

  • The artefact payload (normalised, hash-verified)
  • The admissibility proof (plausibility + probability certifications)
  • The citation record — specifying how the artefact should be cited

An AI system that cites an AIEP artefact can prove:

  1. The artefact existed at time T with hash H
  2. It passed the admissibility gate with certification C
  3. The citation is in canonical form and matches the record

This makes AI-generated citations machine-auditable after the fact.

P60 defines the retrieval schema. P61 defines citation normalisation for AI responses. P62 governs cross-jurisdiction retrieval: resolving artefacts whose legal identity differs across jurisdictions.

Training Data Governance — P63

Training Data →

AIEP-governed training data is:

  • Normalised (deterministic hash for every artefact)
  • Admissibility-gated (the same fail-closed gate applies to training as to inference)
  • Auditable (the full evidence chain for every artefact is committed to the ledger)

P63 governs training data ingest to an AIEP substrate: artefacts that do not pass the admissibility gate cannot be used in AIEP-governed training. The exclusion record is committed. The model’s effective training corpus is a machine-verifiable set.

This is the foundation of AIEP’s “explain the data” compliance posture.

Validator — P64

Validator →

The AIEP validator is the standalone compliance check for AI-generated artefacts. Given:

  • An AI-generated output
  • The claimed evidence chain

The validator verifies:

  1. Every cited artefact exists in the AIEP ledger at the claimed hash
  2. Each artefact passed the admissibility gate with the claimed certification
  3. The divergence proof for any conflicting artefacts is committed
  4. The output’s claimed confidence range is consistent with the Probability Certification

Any AI system that produces AIEP-validated output is producing output that any third party can re-verify independently.

Open-source repositories

RepositorySpecificationsDownload
aiep-retrieval-protocolP60-P62MD spec - zip
aiep-training-governanceP63MD spec - zip
aiep-validatorP64MD spec - zip