Story
September 30, 2026
Trump’s AI Safety Deal Leaves Big Tech Policing Itself
The White House and major AI companies present the accord as a practical first layer of safety that protects innovation; critics see the same deal as a weak endorsement of self-regulation, with the companies setting—and judging—their own limits.
In the run-up to Tuesday’s White House gathering, the AI industry was already split. Some executives had warned that rapidly advancing models required stronger safeguards, while President Donald Trump argued the US could not slow a technology he said would eclipse the Industrial Revolution.1
At the luncheon, Trump brought together leaders including Google’s Sundar Pichai, Meta’s Mark Zuckerberg, Anthropic’s Dario Amodei, OpenAI’s Greg Brockman, xAI’s Elon Musk and Nvidia’s Jensen Huang. They signed the White House Accord on Super Intelligence—a voluntary, two-page commitment built around internal monitoring, dedicated safety teams, independent outside evaluations and board-level oversight.2
Trump called the arrangement “morally binding” and said he was seeing “tremendous self-policing.”1 His administration’s wager is clear: the companies closest to frontier models should bear the primary responsibility for controlling them, rather than face sweeping federal rules. Musk amplified that principle in reposting Huang’s account of the meeting: companies building the technology are responsible for deploying it safely and being accountable.
3
Zuckerberg framed the pact as a confidence-building measure, calling the industry-wide commitment to “robust internal controls and multiple layers of audits and reviews” a “significant positive step.”
4 Yet even supporters cast it as an opening move, not a finished regulatory regime: the accord says its practices may eventually be written into law.2
That caveat is the heart of the criticism. Venture investor Bradley Tusk noted that many of the same chief executives had recently argued for regulation, suggesting common rules would level the competitive field.5 Other critics say the pledge leaves crucial questions unanswered: companies can select and pay evaluators, define whether a model is behaving “as intended,” and keep audit findings from public view.6
The result is a striking compromise. Washington has secured a shared safety vocabulary and promises of review, while leaving enforcement—at least for now—in the hands of the companies racing to build the technology.