Story
September 30, 2026
Trump’s AI Pact Puts Safety in the Hands of the Companies Racing Ahead
Trump’s White House accord asks leading AI companies to police themselves through internal controls, outside reviews and board oversight. Supporters call it a practical start; critics see a voluntary pact with no enforceable public safeguards.
In the weeks before Tuesday’s White House meeting, AI safety had become a sharper fault line. Some executives, including Anthropic’s Dario Amodei, had urged greater caution as concerns mounted over powerful models; Trump, by contrast, framed curbing development as a strategic error in a contest with China.1
At the meeting, Trump gathered leaders from Google, Anthropic, Meta, OpenAI, xAI and Nvidia to sign the White House Accord on Super Intelligence — a voluntary, two-page framework he described as “morally binding.” Its central bargain was clear: the industry would keep moving, while building its own guardrails.2
The accord calls for four layers of oversight: internal controls to monitor risks including cyber, biological and chemical threats; a company team to verify those controls; independent outside evaluators; and a board committee to receive reports and press for fixes. Participating companies also agreed to meet regularly on safety standards, while leaving open the possibility that the practices could someday be written into law.3
Trump hailed “tremendous self-policing” and insisted Washington would not “stifle” a technology he believes will eclipse the Industrial Revolution. He argued that the Justice Department and FBI already provide a backstop if firms stray into dangerous conduct.2
The companies largely embraced the moment. Mark Zuckerberg called the joint commitment a “significant positive step,” saying robust internal controls and multiple layers of auditing should strengthen public confidence.
4 Still, the industry’s unity was qualified: Amodei said the technology carries “very real risks” and that the mechanism for addressing them remained unsettled.5
That caveat is where critics see the accord’s weakness. One detailed assessment argued that undefined terms such as “robust” controls and “operating as intended,” combined with company-chosen evaluators and no public reporting, make the arrangement difficult to test or enforce.6 Sen. Mark Warner made the competing case for law: companies developing the most powerful systems, he said, are warning that the technology is advancing faster than its safeguards.1
For now, Trump has secured a show of corporate alignment, not a regulatory settlement. The real test will be whether voluntary audits uncover problems — and whether companies, or Washington, act when they do.