What is AIEP?
AIEP — Architected Instruction & Evidence Protocol — is an open protocol that links instructions to evidence, enabling organisations to publish structured, verifiable knowledge that both humans and AI systems can retrieve, validate, and trust.
It is the infrastructure layer that the web always needed but was never built: a standard for verified, machine-readable, evidence-bound knowledge.
The problem AIEP solves
The web is the world’s largest knowledge store. But it was never designed for machine verification.
A webpage can say anything. There is no standard way for a retrieval system — human or AI — to check:
- Whether a claim is current (it may have changed since publication)
- Who made the claim (an expert authority or an anonymous assertion)
- Whether the evidence supports it (or whether confidence was mistaken for proof)
- Whether the document has been tampered with (no integrity check exists by default)
AI models are now used as the primary interface for knowledge retrieval. When those models retrieve from the web, they inherit every one of these problems — and then present the result with the same confidence regardless of its evidential quality.
The consequence: AI-retrieved knowledge is structurally unreliable. Not because AI is bad. Because the web was never architected for verification.
What AIEP does
AIEP solves this at the protocol level — not with better AI, but with better infrastructure.
| AIEP capability | What it means in practice |
|---|---|
| Instruction-evidence linking | Every claim carries a machine-readable reference to the artefacts that support it — not a footnote, but a hash-verified field |
| Structured artefact publication | Knowledge is published as schema-conformant JSON under /.well-known/aiep/ — any AI system can retrieve and validate it |
| Machine-readable provenance | Who published it, when, what hash it carried — all encoded in the artefact, not inferred from the page |
| Dissent preservation | Outlier claims are archived with their evidence intact, ready for recall when knowledge shifts |
| Execution-gated reasoning | Before acting on a claim, a conformant AIEP system checks plausibility and probability certification |
| Constitutional enforcement | The GENOME kernel enforces four invariants at the execution substrate — making violation architecturally impossible |
AIEP in plain English
Think of it as the difference between a written report and a sworn affidavit.
Both contain claims. But the affidavit has a named author, a date, a signature, and consequences for falsehood. The claims are linked to their source. The document carries evidence of its own integrity.
AIEP does this for knowledge published on the web — automatically, machine-readably, and verifiably.
A publisher implementing AIEP does not just publish a webpage. They publish:
- The human-readable content (as normal)
- A structured, schema-conformant artefact at a known path
- A cryptographic hash so any retriever can confirm the content has not changed
- A provenance record linking the artefact to its issuer
- An evidence chain showing what the claim is based on
A retrieval system — AI or otherwise — can then verify all five in sequence without contacting the publisher.
What AIEP is not
- Not a new AI model. AIEP is infrastructure, not intelligence. It governs how AI systems retrieve and verify knowledge.
- Not a content platform. AIEP does not host your content. It defines how you publish it.
- Not surveillance or gatekeeping. Anyone can publish. AIEP adds a verifiable layer on top of open publication.
- Not a replacement for the web. AIEP is complementary. Existing web content continues to function. AIEP adds what was always missing: a verification layer.
- Not proprietary. AIEP is open. Open use is always permitted under Apache 2.0. No registration. No fee.
The components
GENOME SDK — The Python reference implementation. Enforces constitutional invariants at the execution substrate. Available on PyPI.
Mirror — The machine-readable publication surface. Every conformant publisher hosts a Mirror at /.well-known/aiep/.
Piea — AIEP in production. An evidence-bound AI assistant where every response is cryptographically committed and replayable.
Hub — The public developer gateway. Schemas, specs, downloads, and the machine interface.
How to get started
- Read the specification — /spec — normative protocol requirements
- Explore the schemas — /schemas — canonical JSON schemas for every artefact type
- Explore the SDK — /genome-sdk — the Python reference implementation of the AIEP constitutional stack
- Publish a Mirror — host
index.jsonandmetadata.jsonat/.well-known/aiep/on your domain - Download artefacts — /downloads — whitepapers, specs, and reference bundles
See also: About · Explainer · Architecture · Protocol · Get Started