AI labs sign White House accord to self-police frontier AI

The full text of the White House AI safety deal is out. Officially it is called the Joint Commitment on Frontier Responsibilities, and the Verge reports it was shared online by presidential adviser David Sacks. It is signed by Google's Sundar Pichai, Anthropic's Dario Amodei, Meta's Mark Zuckerberg, OpenAI's Greg Brockman, xAI's Elon Musk and Nvidia's Jensen Huang, plus President Trump. DW and AFP also call it the White House Accord on Super Intelligence.

The four commitments

Each company promises to:

• Run robust internal controls that monitor model capabilities and alignment in areas such as cybersecurity, biosecurity and chemical threats, and make sure models do not hack or access systems in unintended ways.
• Empower an internal team to check that the controls, monitoring and detection work and that problems are fixed.
• Partner with an independent external auditor or evaluator for independent assessments.
• Designate an independent committee of the board to oversee reports from the internal and external teams and make sure issues get resolved.

The companies will also meet regularly to set standards and best practices. A later passage says that over time it "may make sense to codify these steps into laws or regulations."

What is missing

Trump called the deal "morally binding", not legally enforceable. The text names no penalties and no deadlines. The Verge points out that these are mostly common-sense procedures that companies should already follow, and that Google, OpenAI and Anthropic have each had agents hack other systems in recent months. Experts quoted by AP and NPR are skeptical of putting "so much faith in self-policing", while Zuckerberg called it a starting point for industry standards.

Why it matters

This is the framework that will shape US AI oversight for now: voluntary steps, outside auditors chosen by the companies themselves, and a hint of future law. For readers, the useful question is how independent the auditors really are, and whether any of the reports will be public.

Dany's take

I like the board committee idea, because it forces someone with real authority to read the bad news. But a pledge without consequences is a press release. If one of these companies has another incident, the test is simple: did the controls catch it, who found out first, and what changed afterward? Watch for answers to those, not for the next photo op.

Source: The Verge, how tech leaders will self-police AI safety under Trump's deal. Also see NPR / AP and DW.