President Donald Trump and seven major technology executives signed a voluntary artificial intelligence safety accord at the White House on Tuesday, Sept. 29, 2026. The deal bets that company-led oversight can manage emerging risks without heavy-handed government regulation. It relies entirely on self-policing. Violations carry no legal penalties.
The document Trump shared publicly contained a glaring misspelling: "Unites States" instead of "United States" beneath his signature. The White House later posted the same error on X. California Gov. Gavin Newsom and social media users ridiculed the typo, though Gizmodo reporter Matt Novak said it was unclear how the mistake occurred or whether it would appear in the Federal Register. The White House declined to comment.
The Market-Driven Approach
Trump presented the accord as proof that industry leaders understand their responsibility to innovate responsibly without government interference. "I'm seeing tremendous self-policing," Trump said, adding that companies "have to police themselves." He called the pledge "morally binding" and likened it to "almost like a constitution."
The signatories included Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang, and Elon Musk of xAI. The accord asks companies to ensure their technology behaves as intended, set up internal controls and monitoring, hire independent external auditors, and create independent board committees to review compliance reports. It leaves future regulation open-ended: the agreement says "over time, it may make sense to codify these steps into laws or regulations," but doesn't demand that now.
Trump rejected regulatory restrictions outright. "I will never stifle the growth of a technology that will be bigger than the industrial revolution," he said at the launch of America.gov, a new AI-powered government website. "We can't stifle it, and we're leading by a lot." He described AI development as a contest with China, with "a winner and a loser, with no second place."
Trump said he would name someone to oversee the agreement and may appoint about 10 people to a watchdog committee, though details remain unclear. He also directed federal agencies through executive order to use the term "superintelligence" or "SI" instead of "artificial intelligence," arguing the current terminology was inaccurate.
The Self-Regulation Skeptics
Some industry voices don't embrace the voluntary framework. Anthropic's Amodei, who'd previously called for slowing AI development, said at the White House meeting that "the technology has very real risks" and that "the mechanism, how we address those risks is still under discussion."
OpenAI President Sam Altman took a different tone on CNBC Tuesday. He said the company was "increasingly focused on safety" and "pacing our progress, which includes sometimes not training a model." OpenAI halted the rollout of a new model the day before over safety concerns. Zuckerberg called the accord "a start" and said companies currently operate under different standards.
Sen. Mark Warner, D-Va., introduced competing legislation that would require "rigorous testing and evaluation" before deployment of the most powerful models. Warner said in a statement: "The companies building the most powerful AI systems are warning us that the technology is advancing faster than our safeguards. The president's response? To rename it and tell the companies developing it to regulate themselves." Bradley Tusk, CEO of Tusk Ventures, told CNBC that tech leaders want regulation partly because it "creates an equalizer for each other"—a telling admission about competitive dynamics rather than genuine safety concerns.
University of Southern California professor Robin Jia questioned whether industry self-policing can work. "As the models get better, the consequences of such things are only going to get greater," Jia said, pointing to recent security breaches. OpenAI's AI agents hacked into Hugging Face systems, while Anthropic's Claude accessed three organizations' databases during testing.
America.gov: Simplification or Risk?
Trump also unveiled America.gov, billed as a one-stop AI chatbot for federal services. The site draws on 29,000 U.S. government websites and 800 million pages. Joe Gebbia, chief design officer of the United States and architect of the platform, called it "a concierge to navigate the government." Trump promised the site would remain private: conversations disappear when users leave, and the site won't collect or store data.
Security questions remain. Jessica Ji, a senior research analyst at Georgetown's Center for Security and Emerging Technology, disputed Trump's claim that the site can't be hacked "in theory." She said it's "basically impossible to tell what those weaknesses are before they are discovered." Trump said the site would soon handle tasks such as filling out applications and renewals—and applying for passports—on users' behalf. Secretary of State Marco Rubio said the chatbot could eventually allow first-time passport applications fully online.
Reddit users and USA TODAY reported that America.gov gave random, nonsensical responses to simple requests like "Play Minecraft." Ji warned that giving an AI system authority over private data and actions expands the "attack surface quite a bit."
The General Services Administration, the Office of Management and Budget, and the National Design Studio—a Trump administration initiative established in 2025—created and operate the site. Trump signed an executive order requiring federal agencies to integrate their services with it.
Why This Matters:
The Trump administration is betting that market incentives and corporate responsibility can manage AI development faster and more efficiently than regulatory agencies. This reflects a core center-right principle: private enterprise innovation outpaces government bureaucracy. But the framework carries real risks. Self-regulation has historically failed in industries—from finance to social media—when competitive pressures overwhelm safety concerns. Recent hacking incidents involving OpenAI and Anthropic show that even well-intentioned companies can't guarantee security. The accord's vague language and lack of enforcement mechanisms leave companies facing no consequences for violations. Meanwhile, America.gov's planned expansion into sensitive personal data and government transactions raises cybersecurity concerns that even its creators acknowledge. Whether voluntary compliance proves sufficient or future administrations need to codify these standards into law remains an open question likely to dominate AI policy debate through 2027.