How AIEP Works

AIEP adds a verification layer to AI outputs. The core mechanism has four steps: retrieve evidence, hash it, bind it to the response, validate.

Step 1 — Evidence retrieval

When a question is submitted, the system retrieves relevant source artefacts from the web or a configured evidence store.

Each artefact is a source document: a URL, a document, a structured JSON record, or a schema file.

Step 2 — Hashing

Each retrieved artefact is hashed using SHA-256. The hash is a fixed-length fingerprint of the document content at the moment of retrieval.

Source content → SHA-256 → content_hash: "f3a2c4e0b6a7d991..."

If the source changes after this point, a replay will produce a different hash. The discrepancy is detectable.

Step 3 — Evidence rail construction

The hashed artefacts are assembled into an evidence rail — an ordered, machine-readable list of everything the response was based on.

{
  "evidence_rail": [
    {
      "artefact_id": "ev_001",
      "source_url": "https://aiep.dev/schema/evidence.json",
      "content_hash": "f3a2c4e0b6a7d991...",
      "retrieved_at": "2026-04-25T00:00:00Z",
      "confidence_tier": "verified",
      "validation_status": "passed"
    }
  ]
}

Step 4 — Validation

The complete response — answer text plus evidence rail — is validated against the AIEP canonical schema.

Validation checks:

• Schema conformance
• All artefact hashes present
• Evidence rail non-empty
• Replay metadata present

If validation passes, the response is marked status: "passed" and returned. If it fails, the response is flagged.

Machine-readable publication

AIEP domains publish their evidence infrastructure at a known path:

/.well-known/aiep/index.json      — protocol discovery
/.well-known/aiep/metadata.json   — domain identity and version

Any crawler, AI system, or auditor can retrieve these endpoints without authentication.

Public repositories

The full AIEP implementation is open source:

Try it

Verification Playground → — run through all four steps interactively

View a verified response →

View the response JSON →

Machine endpoint →

See also: What is AIEP? · RAG vs AIEP · AI Audit Trail · Deterministic AI · Build with AIEP