Six AI companies sign White House audit agreement; commentary points to flaw of companies choosing their own auditors

On October 17, according to a commentary published by CoinDesk, Samuel Vance-Law, research director at the Decentralization Research Center, pointed out that the White House frontier AI agreement signed on September 29 by Nvidia, OpenAI, Anthropic, Google, Meta, and xAI (which Trump called "morally binding") has a key flaw: the agreement does not specify who selects the auditor, what can be reviewed, or whether the findings will be made public, and it carries no penalties for breach. In practice, each company chooses its own auditor and determines the scope of review. He cited the example of Enron hiring Arthur Andersen as its auditor in 2000, which later gave rise to the Sarbanes-Oxley Act, and referenced an assessment showing that 16 companies fulfilled an average of only 53% of their 2023 voluntary White House commitments, with 8 scoring zero on assisting with external error detection. He argued that auditors should be able to independently choose what to test, that companies should disclose their auditors and the scope of review, and that oversight should be handled by an institution independent of the companies.