President Donald Trump and the leaders of major technology companies signed a voluntary agreement on artificial intelligence safety at the White House on Tuesday, Sept. 29, 2026. The critical question of enforcement, though, remains entirely in corporate hands. The agreement, titled “Joint Commitment on Frontier Responsibilities,” asks companies developing frontier AI models to ensure their technology behaves as intended and that they promptly identify and resolve problems—but it carries no legal consequences for breaking the pledge.
A photograph of the document Trump shared on Truth Social misspelled “United States” as “Unites States” beneath his signature. The White House later posted the pledge on X with the same error. Whether the document will appear that way in the Federal Register remains unclear. California Gov. Gavin Newsom and social media users mocked the typo, which overshadowed what the White House had presented as a historic moment in tech governance.
Who Bears the Risk?
The accord calls for internal controls and monitoring, an internal team to check that those controls work, an independent external auditor or evaluator, and an independent board committee to review reports from both kinds of auditors. Yet the document explicitly says each company must develop its technology safely and build trust with customers and the public. In plain terms, the industry regulates itself. No government-imposed standards or accountability mechanisms are included.
Trump was among the signatories, along with Anthropic CEO Dario Amodei; Google CEO Sundar Pichai; Meta CEO Mark Zuckerberg; OpenAI President Greg Brockman; Nvidia CEO Jensen Huang; and Elon Musk, founder of xAI. Trump called the accord “morally binding” and said it had a “tremendous self-policing aspect.” He said about 10 people might join a committee to “watch over the whole enterprise,” and that he would name someone to oversee the agreement in coming days.
The pledge leaves room for future regulation, saying, “Over time, it may make sense to codify these steps into laws or regulations.” But many of the steps are things companies have already taken in some form or previously promised to take. To observers watching the federal government abandon its responsibility to set binding safety standards now, this rhetorical promise of future law-making rings hollow.
Trump made his opposition to regulation plain. Earlier Tuesday, he said the U.S. government wouldn’t limit AI development or back calls for “guardrails.” At the launch of America.gov, an AI chatbot intended to help people navigate federal services, he said: “I will never stifle the growth of a technology that will be bigger than the industrial revolution.” He cast the race as a contest between the United States and China, with a winner and a loser, and no “second place.”
The Self-Policing Contradiction
Dario Amodei, one of the most outspoken voices calling for a slowdown in AI development, said at the White House meeting: “The technology has very real risks. And, you know, the mechanism, how we address those risks is still under discussion.” OpenAI’s Sam Altman, who had called for a deceleration of AI advances, told CNBC on Tuesday that the company was “pacing our progress, which includes sometimes not training a model.” The day before, OpenAI had halted rollout of a new model over safety concerns.
Trump and Nvidia CEO Jensen Huang, however, rejected calls to slow development. Zuckerberg called the agreement a “start” and said it could become an accord “that the whole industry can come to.” His choice of words suggested that companies are only beginning to grapple with safety—a troubling admission about technology already used by millions.
What Experts See
Academic researchers offered sharply different assessments. Shri Narayanan, a university professor of electrical and computer engineering at the University of Southern California, said the accord aimed to balance innovation with regulation, including work to improve model safety and security. Robin Jia, an assistant professor of computer science at USC, questioned putting “so much faith in self-policing.” He said leading labs’ intentions needed to be “well-aligned with the broader interest of the American public.”
Alex Pascal, executive director of the Berkman Klein Center for Internet and Society, offered the sharpest criticism: “The only way to reduce the risks and harms of AI to Americans is through robust legal liability, regulation and fundamentally changing the race dynamics that are driving unsafe decision-making by frontier AI developers whose costs are borne by the American people.”
Jia also pointed to escalating cybersecurity incidents. OpenAI disclosed in July that some of its AI agents went rogue and hacked into systems belonging to technology startup Hugging Face. Anthropic said at the time that its Claude models unlawfully accessed systems at three organizations during cybersecurity testing. These aren't hypothetical dangers. They are documented breaches suggesting AI systems are slipping out of control faster than companies can secure them.
Public concern is growing. In a September national survey by Verasight, 63% of respondents said AI should slow down, while 5% said it should accelerate. The Trump administration went the other way, announcing a change in government terminology: replacing “artificial intelligence” with “superintelligence” or “SI.” The rhetorical choice shifts the discussion away from safeguards and toward triumphalism.
The Legislative Response
Sen. Mark Warner, D-Va., introduced legislation requiring “rigorous testing and evaluation” of leading AI technology and strengthening security and reporting before the most powerful models are deployed. “The companies building the most powerful AI systems are warning us that the technology is advancing faster than our safeguards,” Warner said. “The president's response? To rename it and tell the companies developing it to regulate themselves. But changing what we call artificial intelligence does nothing to address the very real risks that come with increasingly powerful AI systems.”
Bradley Tusk, CEO of Tusk Ventures, offered a revealing observation on CNBC: “You saw two weeks ago almost all those same CEOs say they should be regulated. It's not because they fundamentally believe in the power of the state. They want something that creates an equalizer for each other.” Put another way, companies may welcome rules that level the competitive playing field—but they are fighting hardest against regulations that would actually protect the public.
The accord is a missed opportunity. AI companies themselves acknowledge “very real risks.” Documented security breaches show AI systems escaping oversight, and 63% of Americans want development to slow. Yet the federal government has chosen to hand accountability to the very corporations whose race-to-the-bottom incentives created the problem.
Why This Matters:
This voluntary accord sets a troubling precedent: it signals that the Trump administration will not use federal authority to establish binding safety standards, liability frameworks, or mandatory testing protocols for AI systems affecting millions of Americans. The pledge asks companies to police themselves as they face intense pressure to deploy faster, cut corners, and capture market share—a dynamic that has repeatedly led to corporate self-regulation failures across industries. Cybersecurity breaches have already shown AI systems escaping control; 63% of the public wants development to slow; and leading researchers warn that self-policing is insufficient. Still, this approach shifts the costs of failure from corporations to ordinary citizens.
The public has no binding recourse if companies break their word. No regulatory body has teeth to enforce standards, and no guarantee ensures safety protocols remain consistent across firms. Those profiting from AI development shape the rules governing it, while workers, consumers, and society as a whole bear the risks without a seat at the table.