AI Proof Audit
Two weeks to find out what your AI cannot prove.
A fixed-scope audit of one AI system: what it does, what it may act on, what evidence exists, and where an independent review would fail. You get a ranked gap list, not a framework deck.
Most AI programmes fail review for the same reasons. The system cannot show what it retrieved, which identity acted, or which test would catch a regression. The model is rarely the weak point.
We audit one system at a time. Interviews, a read of the code and configuration, and a run against your own corpus. No generic maturity questionnaire, no scorecard that could describe any company.
The output is deliberately boring: a ranked list, each item with an owner and a test that proves it is fixed. If the gaps are small, the audit says so and you keep the difference.
Who it is for
Teams with a pilot heading into review, or a system already in production that nobody can fully explain.
Investment
Fixed price agreed before work starts. Credited toward a build if you continue within 90 days.