Story
September 30, 2026
Trump’s AI Accord Bets on Self-Policing as Safety Fears Mount
The White House and technology executives argue that industry-led checks can move faster than Washington, while safety-minded critics and lawmakers say voluntary promises leave the public exposed when Congress is still struggling to act.
Before Tuesday’s White House gathering, Congress was already confronting an uncomfortable contradiction: companies say frontier AI could carry catastrophic risks, yet the technology is also becoming a major engine of investment and economic growth. Speaker Mike Johnson argued the answer should be a light touch, saying he hoped guardrails would be “voluntary” and that companies wanted to reassure investors, consumers and the public that they could operate safely.1
That stance put the administration on a collision course with those demanding enforceable safeguards. Sen. Mark Warner warned that an AI incident involving national security or critical infrastructure could arrive while Congress is away; he planned to seek fast-track passage of a bill requiring a federal safety-board review of frontier models before public release. The proposal was considered unlikely to pass, underscoring the legislative vacuum.1
At the White House, Trump and Johnson convened a high-powered lunch of officials and executives, including leaders from Meta, Nvidia, Google and Anthropic, alongside Elon Musk, Jeff Bezos and Satya Nadella. The administration’s expected hands-off approach was clear: rely on the companies’ own technical expertise to set the guardrails, with few immediate regulatory specifics.2
The resulting voluntary accord calls for transparency, internal risk reviews and external audits. Mark Zuckerberg described “multiple layers of auditing controls,” while Trump called the agreement “morally binding” and said the companies would be “policing each other.” But people working in AI told Axios that self-policing would not resolve the industry’s safety problem, particularly when the deal’s scope and enforcement have not been fully disclosed.3
Trump also pressed a rebrand, calling AI “SI,” for superintelligence. Musk amplified the new language online, reposting the White House accord as a “Joint Commitment on Frontier Responsibilities.”
4 The slogan may be settled; the question of who enforces the commitments is not.