How the terms relate
VERIFIABLE AI · A NEW CATEGORY OF AI
OPEN VERIFICATION · A NEW CATEGORY OF OPEN SOURCE
Open verification is a new category of open source: the open standards and specifications that let anyone verify what an AI system did, without trusting the operator or an auditor. It is the foundation. Verifiable AI is a new category of AI: AI whose actions can be verified from evidence, built on open verification. Proof-of-Control is our flagship standard for evidence that an AI agent stayed within the controls it was given. It is one of many ways to achieve open verification.
Claims-based AI and verifiable AI
Claims-based AI asks you to trust agent logs. Verifiable AI produces a separate piece of evidence of what the agent did, which is either attested by an independent verifier or, with open verification, verifiable by anyone without trusting the operator or an auditor.
AI Tomorrow
| Claims-based AIVerifiability Tier 1 · where most AI is today | Verifiable AIFrom Verifiability Tier 2 · at minimum | Verifiable AI with open verificationProof-of-Control Verifiability Tiers 3 and 4 | |
|---|---|---|---|
| Who is the actor? | Inferred from credentials. | Attested. | The actor's identity is bound to a known operator, and anyone can openly verify it. |
| What were they allowed to do? | Asserted in policy. | Authority is documented and signed in policy. | Authority is signed and scoped, and anyone can openly verify each grant. |
| What did they actually do? | Actions reconstructed from logs. | Each action carries evidence. | Each action carries tamper-evident evidence anyone can openly verify. At Verifiability Tier 4, the action cannot run without it. |
| What is the agent made of? | Undocumented, changes silently. | Composition documented and attested. | Composition is fingerprinted, so a swapped model shows up. |
| Who can verify? | Only the vendor. | Someone other than the vendor. | Anyone can openly verify, without trusting the operator or an auditor. |
Verifiability Tiers
VERIFIABILITY TIER 1
Assertion
their word
VERIFIABILITY TIER 2
Attestation
a third party vouches
VERIFIABILITY TIER 3
Trust-minimized
anyone can openly verify
VERIFIABILITY TIER 4
Self-enforcing
no evidence, no action
THE BINARY THRESHOLD: Graded by who you must trust, not by whether cryptography is used. It indicates the phase change from human verification to mechanical verification, from a party’s word to a mechanism anyone can openly verify.
Open verification means anyone can verify what an AI system did, using published methods, without having to trust the operator or an auditor. It is essential when machines act faster than people can watch. Open verification is a new category of open source: the open standards and specifications that let anyone verify what an AI system did, without trusting the operator or an auditor.
Verifiable AI is a new category of AI: AI whose actions can be verified from evidence, built on open verification.
Claims-based AI is AI that asks you to trust what it did on the operator's word, whereas verifiable AI provides evidence of what it did.
Proof-of-Control is our flagship industry-led open standard for evidence that an AI agent stayed within the controls it was given. It is one of many ways to achieve open verification. Anyone can openly verify that evidence without trusting the operator or an auditor, and any change made after it is recorded can be detected.
The widening distance between what AI agents do and anyone's ability to openly verify that they stayed within the controls they were given.
WHAT
AGENTS DO
WHAT ANYONE CAN OPENLY VERIFY
VERIFIABILITY GAP
WIDENS AS CAPABILITY COMPOUNDS
A four-level framework for verifying an agent's control adherence, ranked by trust, from taking the operator's word to evidence anyone can verify without trusting the operator or an auditor.