GLACIS
Platform
Solutions
AI in production Supervise the AI you run, with a signed record others can check AI you sell Show customers how your AI is controlled, in a record they can check AI you oversee Read the signed record from a vendor’s AI instead of taking their word for it All solutions Workflows and industries
Standards
Operational evidence How signed records become evidence a reviewer can check Sample record Every action, from its rule to a record anyone can check OVERT standard The open record format, published June 2026 Verify a record Check a signed operational record yourself, in your browser
Resources Company Talk to us Start free
GLACIS

Navigate

Home PlatformRules, control decisions, and records anyone can verify Resources Company

Solutions

AI in productionSupervise the AI you run, with a signed record others can check AI you sellShow customers how your AI is controlled, in a record they can check AI you overseeRead the signed record from a vendor’s AI instead of taking their word for it All solutionsWorkflows and industries

Standards

Operational evidenceHow signed records become evidence a reviewer can check Sample recordEvery action, from its rule to a record anyone can check OVERT standardThe open record format, published June 2026 Verify a recordCheck a signed operational record yourself, in your browser
Start free Talk to us

Operational evidence · OVERT

What Makes AI Attestation Independently Verifiable?

A signed record can let a third party check specific claims about a covered event. It does not, by itself, prove control effectiveness, complete coverage, safety, data locality, or the signer’s organizational independence.

Joe Braidwood
Joe BraidwoodCo-founder & CEO
Updated August 2026 · 6 min read

AI systems increasingly take actions that someone may later need to explain: allowing a transaction, denying a request, escalating a case, calling a tool, or releasing an output. A policy states what was supposed to happen. Operational evidence helps establish what a particular system component reported doing for a particular covered event, which is a narrower thing to say and an easier thing to check.

The distinction survives only if the language around it stays careful. “Independently verifiable” means a reviewer can apply a disclosed verification procedure without relying on a proprietary dashboard. It does not mean the signer, witness, or evidence producer is necessarily independent of the operator. It also does not turn a signed claim into proof that the underlying statement is true.

For a primer on the mechanics, see AI attestation. This guide focuses on the limits and value of third-party verification, using the OVERT open standard as the reference format.

What independent verification means

A record is independently verifiable when its format, covered fields, signatures, and checking procedure are sufficiently disclosed for another party to perform the check. A successful check can establish that the signed bytes have not changed and that the signature corresponds to the stated key. Depending on the schema and evidence supplied, it may also support reconstruction of the reported control decision and event sequence.

Those checks do not establish everything a buyer, auditor, insurer, or regulator may care about, and that shortfall is where most arguments start. Four questions have to stay separate:

  • Integrity: do the record and signature still match?
  • Attribution: which key signed which covered fields?
  • Truth: was the signed claim factually accurate?
  • Coverage: which actions entered the evidenced path, and which did not?

Cryptographic verification principally addresses integrity and key attribution. Truth, control effectiveness, completeness, and organizational independence require additional evidence.

Four properties of a useful attestation record

A bounded claim

The record should say exactly what is being claimed: the configured control, the reported decision, the event identifier, the covered fields, and the relevant time or sequence. “This component reported a deny for this covered event” is testable. “The AI was safe” is not.

Configured data minimization

An implementation can keep protected payloads local while exporting hashes, metadata, signatures, or verification material. That is a deployment property, not something a receipt format proves automatically. If a remote model or tool endpoint receives content, the content has still crossed that boundary.

Disclosed signer and witness provenance

A verifier should be able to see which keys signed the record and what role each signer claims. A service-operated witness, countersignature, or inclusion proof can make later alteration more detectable. A second signature does not by itself establish that the witness is organizationally independent, that the original event was captured correctly, or that every relevant action was recorded.

Stated scope and measurement

Coverage claims need a denominator: the population of actions expected to traverse the configured path, the exclusions, the observation window, and the method used to count them. Without that context, a collection of valid signatures can still leave material coverage gaps.

What an OVERT record can contain

OVERT defines a portable structure for a ControlAction record. Depending on the implementation and profile, a record can identify the policy or control reference, report an allow, block, or escalation decision, identify the signer, link related events, and commit to selected fields with hashes and signatures.

A verifier can check supported signatures and covered fields against the published procedure. The record alone does not prove trusted execution, model safety, legal compliance, complete event capture, the effectiveness of the configured control, or the organizational independence of a signer. Those are separate claims and should be supported separately.

Why an open format matters

An open format reduces dependence on one vendor’s interface. A customer, auditor, insurer, assurance body, or regulator can retain the artifact and apply the same published checks with its own tooling. That portability makes the evidence easier to inspect and challenge; it does not make every claim inside the record true.

This is the useful connection between governance and operation. Governance sets the intended rule. The system applies a configured control to an in-scope action. A signed record can preserve a bounded account of the reported outcome. Independent verification then lets another party check the artifact without confusing that check with certification or an assurance opinion.

The short version

Independently verifiable attestation is a property of the artifact and verification procedure. Organizational independence is a property of the people and entities operating, signing, witnessing, and assessing it. Good evidence makes both visible and never treats one as proof of the other.

Read the published format at overt.is, or Verify a record to inspect a supported signed artifact. To discuss evidence for a consequential AI workflow, Talk to us.

Related research

  • Documentation versus operational evidence
  • Governance tools and operational proof
  • The case for verifiable AI
  • See a sample evidence pack
GLACIS logo GLACIS
Platform Solutions Standards Resources Company Careers
[email protected] Start free Talk to us

© 2026 Glacis Technologies, Inc.

Terms Privacy Cookies Do Not Sell or Share Security Trust Center

We use first-party analytics and, where allowed, B2B marketing technologies. Details