On 8 September OpenAI announced that an internal system had proved part of a Millennium Prize problem. On 11 September the Clay Mathematics Institute responded: apparently settled, evaluation deliberately unhurried. Its own rules require two years of published, unchallenged scrutiny before any claim becomes a result. As of this month the problem remains officially active. The claim is weeks old. The record is not due for two years.
A claim and a record are not the same event, separated only by time.
Set beside most AI governance practice inside organisations today, this is an unusually clean example of a structure that is otherwise rare: a named external body, with no stake in the claim, whose acceptance is required before the claim counts as fact, operating on a clock the claimant does not control.
Most boards evaluating a vendor's claim that a model has solved, verified, or certified something have no equivalent. The claim arrives in a press release or a sales deck. The organisation's own record of having accepted it is dated to whenever someone in procurement or risk signed off, often the same week the claim was made. There is no named external reviewer. There is no published, unchallenged standing period. There is no status that remains “under review” until an independent party says otherwise.
Figure — two clocks, one claim
A test narrow enough to apply this quarter
The test that follows from this can be applied to a specific vendor claim before the end of the quarter. It has three parts.
Identity. Is the claim attributable to the system alone, or does it rest on prior work belonging to named individuals or organisations not credited in the claim as presented?
The record. Has the claim been submitted anywhere capable of refuting it: an independent audit, a named technical reviewer, a certification body? Or does it exist only as material the vendor itself produced and controls?
The anchor. Whose clock is actually running: the date the vendor chose to announce, or the date an external party with no stake in the outcome was satisfied the claim holds?
Most AI capability and compliance claims reaching a board have never been asked to pass this test, because most organisations have never built the equivalent of the Clay Institute's two-year clock into how they receive a vendor's claim in the first place. That absence is not a detail. It is the governance gap.
Locating the gap in a specific organisation
Identifying where that gap sits in a specific organisation, and what a working equivalent of an external, dated review clock would need to look like for the claims that actually reach the board, is the starting point of our advisory work.
A Governance Classification Briefing locates a firm's current position and identifies where the evidential gaps sit before a vendor's claim is tested by someone else.
Book a Consultation