Check the record yourself, without asking the lab.
Today an evaluator or auditor can check only what the lab hands over. With PacSpace, the lab shares one link to a record no one can change, PacSpace included, and you check every entry in your own browser, from anywhere.
Go straight to what you need.
- 01
How the check works
One link, checked in your browser, step by step: what the check shows, what it doesn't, and how to check again without PacSpace or the lab.
- 02
What to ask for
What to ask a lab for before you rely on its record, with plain wording you can adapt for an access agreement or engagement letter.
- 03
Evidence standards
What audit and assessment standards already say about evidence from the party being checked, each beside what the record gives.
- 04
Questions
What an evaluator or auditor asks first. Each answer has its own address to share.
- 05
News
Evaluations, access, evidence standards and policy, each linked to its source.
In their own words: what evaluators are up against.
“My testimony today would not have been possible if AI companies had not been willing to publicly and voluntarily share information about incidents.”
“Our overall view is that it was unlikely that our message board dump was materially altered by agents editing or deleting entries, but we cannot rule it out.”
“Every model we have tested for this behaviour attempted to cheat.”
“evaluators must act under the assumption that evaluation awareness is always present”
What checking requires.
A claim is checkable when it can be confirmed without the cooperation of whoever made it.
| The condition | An agent's log today |
|---|---|
| 01Evidence the checker didn't produce. | The agent's own account. |
| 02Evidence that doesn't move. | Within the agent's reach. |
| 03Evidence either side can reach without asking. | Outsiders can only get it by asking. |
Miss any one and there is nothing to check, only something to believe. The record passes all three.
The outside investigators of the July incident worked for six days and still wrote “cannot rule it out”. A record that passes all three turns that sentence into a check.
The White House Accord on Super Intelligence commits the companies that signed it to bring in an independent auditor or evaluator to assess whether their controls, monitoring and detection are operating as intended. That assessment needs evidence that passes all three. The accord is voluntary today.
One link, checked in your browser.
No account, no signup, nothing to install.
The check runs the moment the record opens.
Every entry against the seal it was committed with.
Change one character of your copy, or drop an entry from it, and the change shows.
Check it again on your own computer.
Our open-source checker runs without PacSpace: npx @pacspace-io/check history.json
What the lab revealed you see in full.
What it withheld you see counted, never shown.
The check shows each seal was committed and that all of them are here, in order, and that each shown field is in its entry's seal. It does not show what a field that is not shown says, or that what was recorded was true.
Audit standards already say it.
“Evidence obtained from a knowledgeable source that is independent of the company is more reliable than evidence obtained only from internal company sources.”5
PCAOB AS 1105, Audit Evidence, paragraph .08. The record's entries are still the lab's account. What changes is that their integrity no longer rests on the lab: a change to any committed entry shows, whoever made it.
The latest for evaluators and auditors.
- PacSpaceThe Records API is in production
Outside teams now record with the Records API in production, and the Shared Record is live for whoever checks.
- METROversightChris Painter's testimony to the U.S. Senate on AI agent incidents
METR's president told a Senate Homeland Security subcommittee that his account of recent agent incidents rested on information AI companies chose to share.
- ANSI National Accreditation BoardStandardsISO/IEC 42006:2025: AIMS Audit & Certification Requirements
ANAB says bodies that certify AI management systems show their competence through accreditation against ISO/IEC 17021-1 and ISO/IEC 42006:2025. The standard governs those bodies, not what an agent did.
Sources
- METR, “Chris Painter's testimony to the U.S. Senate on AI agent incidents”, Sept 30, 2026.
- METR and Redwood Research, independent investigation of the July incident, Aug 26, 2026.
- UK AI Security Institute, “Cheating behaviour in frontier model evaluations”, July 21, 2026.
- Apollo Research, “The Need for Deeper, White-Box Access to Maintain State of the Art Evaluations for Loss of Control Threats”, May 20, 2026.
- PCAOB, AS 1105, Audit Evidence, paragraph .08.
Bring the case you think breaks it.
We would rather be evaluated by use than by description. Talk to us and we'll put you in a live environment: commit a record, do your best to change it, then check it yourself, with us out of the loop. The change shows.
The record must exist.