Vision
The founding insight
Every significant failure in modern organisations — in construction, medicine, finance, governance, and AI — shares a structural flaw: a decision was made, but the evidence behind it was not preserved with the decision.
The instruction exists. The outcome exists. The reasoning that connected them — the evidence that justified the instruction at the time it was given — is gone. Reconstructing it is expensive, incomplete, and contested.
AIEP exists to fix this at the protocol level. Not for one organisation. For the web.
What the web currently is
The web is a publishing system. At its core, it answers one question: where is this document?
URLs point to pages. Search engines index them. AI systems retrieve them. The web is very good at finding content.
What it cannot answer: is this document true, current, and backed by evidence?
What the web should be
AIEP’s vision is a web where every published claim answers four questions automatically — without a human intermediary:
| Question | How AIEP answers it |
|---|---|
| Who published this? | Signed issuer identity cryptographically linked to a registered publisher |
| Has this changed? | Hash binding — any alteration after publication breaks the hash |
| What evidence supports it? | Instruction-evidence binding — evidence is part of the artefact, not a footnote |
| Is it still current? | Temporal validity — expiry and validity windows are machine-readable fields |
These are not new questions. Humans have always asked them about documents. AIEP makes them machine-answerable — for the first time, at protocol level, across the open web.
The AI imperative
The timing is not accidental.
AI systems are being deployed at scale to retrieve, reason over, and act on web content. The quality of those actions depends entirely on the quality of what they retrieve. The web was not built for machine consumption — it was built for human reading.
The result is a structural mismatch:
| What AI needs | What the current web provides |
|---|---|
| Verified current knowledge | Pages that may be outdated, altered, or fabricated |
| Evidence-backed claims | Assertions with no machine-readable evidence link |
| Issuer identity | Anonymous or pseudonymous authorship |
| Temporal validity | No expiry, no validity window, no staleness signal |
| Provenance chain | No chain of custody — just a URL that may change |
| Fail-closed execution | No mechanism to block action on unverifiable claims |
AIEP does not replace AI models. It provides the evidence layer that models currently lack. A model retrieving from AIEP-structured knowledge knows not just what is claimed — it knows who claimed it, when, with what evidence, and whether it is still operative.
The recall vision
The most important part of the AIEP vision is not verification. It is recall.
Every knowledge system in history has discarded what did not fit the current consensus. Wrong theories were deleted. Minority positions were suppressed. Evidence contradicting the dominant view was treated as noise.
Repeatedly, throughout history, the discarded view turned out to be correct. Continental drift. Germ theory. Prions. Stomach ulcers caused by bacteria. Retroviruses. In each case, the evidence existed — it simply was not preserved in a form that could be retrieved and reconsidered.
AIEP preserves everything — not because everything is true, but because we do not yet know what will be true.
The dissent archive stores what does not currently fit. The recall mechanism surfaces it automatically when new evidence changes the picture. A claim does not become permanently wrong. It becomes recall-eligible.
| Knowledge state | What it means | What happens to it |
|---|---|---|
CONSENSUS | Supported by current evidence above threshold | Fully operative — retrievable and executable |
OUTLIER | Below consensus threshold — challenged or decayed | Preserved in dissent archive with full evidence chain |
RECALLED | Previously outlier — restored by new corroborating evidence | Re-enters operative state with recall record attached |
SUPERSEDED | Explicitly replaced by a later artefact from the same issuer | Preserved but marked inactive |
This is not a technical feature. It is a philosophical commitment about how knowledge grows.
The network effect
AIEP’s value grows with adoption. One conformant publisher is useful. A thousand is transformative.
When all organisations in a regulated sector publish AIEP-conformant artefacts:
- Regulators query across the entire sector from a single endpoint
- Auditors verify compliance chains without requesting documents
- AI systems retrieve current authoritative knowledge from the source, not from a training snapshot
- Disputes are resolved against tamper-evident records rather than contested memories
The network requires no central coordination. It requires only that each publisher commits to the same eight canonical primitives. The Canon is the coordination mechanism — the shared commitment that makes any publisher’s artefacts verifiable by any other publisher’s tools.
The long-term position
AIEP is designed to be the evidence layer of the web — as foundational as HTTP, as open as TCP/IP, as verifiable as cryptographic signatures.
Not owned by one organisation. Not controlled by one registry authority. Not restricted by geography or jurisdiction.
A protocol that any publisher can implement, any retriever can use, and any auditor can verify — permanently, deterministically, and without permission.
The goal is not to own the standard. It is to establish it, open it, and let the ecosystem build on it.
Knowledge grows when shared.
Related
- About AIEP — what AIEP is in practice
- Why now — the regulatory and technical moment
- The AIEP Canon — the eight primitives the web has been missing
- Architecture of Knowing — how the knowledge model works
- If AIEP succeeds — what the world looks like