White House AI accord: the gap is who sets audit scope and sees results — Contested · resolves 2027-03-31

The debate: the accord that the White House and several frontier AI companies reached on 29 September 2026 calls for four layers of controls, one of them an external audit, and the question is whether these pledges can stop real incidents.

Why it matters: if audits cannot see where things actually go wrong, companies, boards and regulators believe they are protected when they are not.

Who is on each side: industry voices, including Mark Zuckerberg and Aaron Levie, read the accord as proof that self-regulation still works, while SecondSource reads the audit scope as undefined and all outside pressure as coming after the fact.

SecondSource’s reading: the accord’s third layer already asks external auditors to judge whether controls operate as designed, so the claim that it only counts control layers and ignores effectiveness does not hold, and we have withdrawn it.

The real gap is twofold: who draws the audit scope and how wide, and who gets to see the results.

The three public AI agent incidents of 2026 (an OpenAI security evaluation, an Anthropic evaluation environment that was wrongly connected to the internet, and a classifier switched off at the UK AI Security Institute) were all coverage failures: the controls existed but were not running in the evaluation environment.

If the audited company draws its own scope, the audit may miss exactly the spot where things break.

The American Presidency Project(UC Santa Barbara)
“Designate an independent committee of the board of…”

At the due date, check three things: (1) whether any signatory (Google, Anthropic, Meta, OpenAI, xAI or Nvidia) has named its external auditor and the audit scope, with that scope explicitly covering evaluation environments and internal research agents; (2) whether there is public reporting that findings by an auditor or a board committee delayed a model release; (3) whether a real incident occurs in which monitoring was on and in scope but an agent bypassed or switched it off.

What would overturn it

If (1) holds, the claim that scope is the gap is withdrawn for that company; if (2) holds, the claim that all outside pressure comes after the fact is withdrawn; if (3) holds, the reading that real failures are mainly about coverage is rewritten as bypass-first, and the focus of auditing shifts to red-teaming.