H2A Protocol¶

H2A is a neutral, open, proposed standard for the consent, attestation, and revocation of AI-generated likeness and AI memory. It answers one question, verifiably: was this allowed?

H2A evidences; it does not enforce. For every act of use it produces a signed, externally anchored record of whether the transmission was conformant — and when it was not, it records that explicitly, so the record is admissible evidence. The check is free and open; conformant implementations may offer commercial guarantees around it.

A proposed open standard, authored by Kilbowie. Working draft — see Governance.

The four objects¶

Grant — authority to use a subject within a bounded scope, revocable at point of use.
Attestation — a signed receipt of one act of use.
Decision Record — the signed conformant / non-conformant transmission decision.
Status List — a static, signed revocation artefact.

Attestationreceipt of one act of use

Decision Recordconformant / non-conformant outcome

Media profileasset-bound, when granted

Memory profileasset-bound, when granted

reference the grant by grant_id

Grantauthority to use a subject within a bounded scope — the object everything else binds to

points at its revocation state by status.uri + status.index

Status Liststatic, signed revocation artefact — one bit per grant

The Grant is the hub: an Attestation, Decision Record, and the Media / Memory profiles each bind back to it by grant_id, while the Grant itself points at its Status List by status.uri and status.index.

Start here¶

How verification works — the check, in order, fail-closed.
Core specification · Conformance · Schemas
Reference verifier: reference/ in the repository (python -m h2a_ref.demo).