The White House AI Accord Is Morally Binding and Legally Empty
Six CEOs signed a safety pledge the day before a federal probe opened. The pledge creates no obligations. The probe does.
On Sept 29 the White House got OpenAI, Anthropic, Google, Meta, xAI, and Nvidia to sign a voluntary Frontier Responsibility accord: internal controls, oversight teams, outside audits, a board committee. Trump called it morally binding. The document says it's not legally binding. Those two facts are the whole story.
The timing is the tell. One day after the photo op, the FTC opened an industry-wide probe into the same companies over agent harm. The pledge and the probe are not in tension, they're a division of labor: politics hands you a ribbon, an agency hands you a subpoena.
For anyone deploying these tools, the accord is decoration. The FTC's consumer-protection authority is the real risk, and California's subpoena plus Florida's injunction bid show the states aren't waiting on Washington either. Self-regulation is what you sign; enforcement is what finds you.
I'd treat the pledge as a press release and the probes as the operating environment. The companies that take the accord seriously as a floor, not a shield, are the ones less likely to appear in the next investigation's footnotes.