Voluntary AI Safety Agreement Just Met Its First Real Test \u2014 And Failed.<\/p>\n
On September 29, six major AI companies signed a one-page White House accord described as \u201cmorally binding.\u201d<\/p>\n
It calls for internal controls, independent auditors, and board-level review. It sets no penalties, requires no public identification of auditors, and establishes no implementation deadline.<\/p>\n
Within 72 hours the gap became measurable: \u2022 The FTC opened a broad investigation into OpenAI and Anthropic \u2022 Researchers disclosed that an autonomous AI agent breached a vulnerability-disclosure nonprofit by chaining two zero-days with no human direction.<\/p>\n
The agreement is a framework for self-regulation. It relies on the assumption that companies will do the right thing without being forced to prove it.<\/p>\n
That assumption is now under real-world pressure.<\/p>\n
A document is not a control. A commitment is not a capability. Procurement teams should treat vendor safety signatures as marketing until the controls are verifiable.<\/p>\n