AI agent verification is the independent determination of whether the governance claims made about an AI agent hold against evidence, issued before a counterparty relies on the agent. Certisyn issues that determination as a sealed, policy-versioned, revocable record against the AI Governance Verified standard (AIGVS), published at the IETF as an individual Internet-Draft and open for review. Every verdict resolves publicly at registry.certisyn.com.
A counterparty deciding whether to transact with an AI agent asks 3 different questions. The market answers the first 2 well. Certisyn answers the third.
| Question | Layer | What answers it |
|---|---|---|
| Who is the agent? | Identity | Agent name services, agent passports and on-chain agent registries establish who an agent is and who stands behind it. |
| What did the agent do? | Runtime evidence | Execution traces and hardware attestation record what an agent did and where it ran. |
| Are the claims about how the agent is governed true? | Governance verdict | Certisyn: an independent, sealed, revocable determination, issued before the transaction. |
Every determination is a record another party can replay. It is bound to the exact policy version it was made under, anchored so its time of issue is fixed, and it names its own gaps.
| Field | What it gives the reader |
|---|---|
| Policy version | The exact rule set in force at issuance, so a later reader can tell whether the standard moved after the verdict. |
| Derivation root | A hash chain over the evidence and the reasoning, independently recomputable from the record. |
| Public anchor | The root committed to a public chain, so the time of issue is fixed by a party with no interest in the outcome. |
| Maturity level | Documented, Operational or Adversarial-ready. A level is earned against evidence. |
| Could not determine | Every claim the evidence did not reach, stated plainly, with the count out of the total claims assessed. |
| Revocation state | Live. Monitoring withdraws the seal on drift. |
AIGVS verifies how an organisation governs its use of AI. The 8 control areas are:
The platform that builds an agent cannot credibly verify it. The marketplace that lists an agent and the operator that profits from it carry the same limit. Certisyn issues determinations and accredits the partners who issue under its standard, and separation of duties between those 2 functions is enforced in the pipeline. The conflict-of-interest gate runs at the point of sale and at issuance, and it refuses where a conflict exists.
These public endpoints need no account. They return the state of a determination, the anchor check, the open registry search and the coverage report.
# Resolve a determination: state, policy version, claims summary, derivation root, anchor
curl https://certisyn.com/api/v1/public/scorecard/{vro_code}
# Check the anchor for a root
curl https://certisyn.com/api/v1/anchor/verify/{root}
# Search the open registry
curl 'https://certisyn.com/api/v1/registry/search?q=acme'
# Read what the engine covers, with denominators
curl https://certisyn.com/api/v1/verification/capabilityCoverage is published with its gaps. The capability endpoint reports what the engine covers, with denominators, and names every register no domain has claimed. Read it before relying on a verdict for a claim type.
AIGVS verifies governance of the use of AI. A determination describes a subject at a point in time under a stated policy version, which is why monitoring is part of the product. A verdict is evidence a decision-maker weighs, and it sits alongside model evaluation and regulatory approval as a separate input.
AI agent verification is the independent determination of whether the governance claims made about an AI agent are true, issued before a counterparty relies on the agent. Certisyn issues it as a sealed, policy-versioned record against the AI Governance Verified standard (AIGVS).
Identity establishes who an agent is. Runtime evidence records what an agent did. Verification determines whether the claims about how the agent is governed are true. A counterparty deciding whether to transact with an agent needs all 3 answers.
The policy version in force at issuance, a derivation root that anyone can recompute from the record, a public anchor that fixes the time of issue, a maturity level, the claims the evidence did not reach, and the current revocation state.
An accredited Issuing Partner reconciles the evidence against AIGVS and seals the determination. A conflict-of-interest gate runs at the point of sale and at issuance, so a party with a hand in building or operating an agent cannot issue its verdict.
Yes. Monitoring withdraws the seal when the subject drifts from the evidence, and the public registry reflects the change immediately.
Every determination resolves over a public endpoint with no account, key or signup. The record carries what a reader needs to recompute the derivation root and to check the anchor without asking Certisyn.
AI Governance Verified is published at the IETF as an individual Internet-Draft, draft-hillier-certisyn-ai-governance-verified, and is open for review.