President Donald Trump said Tuesday that top artificial intelligence executives signed a "morally binding" agreement holding their companies responsible for the safety of their own technology, following a closed-door meeting at the White House.
Trump called the gathering "an extraordinary meeting" of the minds "building America's Golden Age." He took questions from reporters for about 30 minutes alongside the executives after the session ended.
The guest list included leaders from OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google — the largest players in the U.S. AI industry.
Under the agreement, which Trump posted online, companies must ensure their technology is safe, implement safeguards to keep models operating as intended, and quickly detect and fix problems when they arise.
The firms also committed to working with "independent auditors" to assess whether their AI systems perform as designed and to prevent platforms from hacking or accessing technical systems "in unintended ways."
The pledge is not legally enforceable. Trump described it as a "form of protection" against the risks AI poses, though he did not detail how compliance would be monitored or what happens if a company breaks its word.
The meeting comes as tech executives and AI researchers increasingly warn about the dangers of advanced AI systems, including the risk of misuse, unintended behavior and loss of human control. Some in the industry have called for tighter government oversight, while others argue that heavy regulation could slow U.S. competitiveness against China.
Trump's approach so far leans on voluntary commitments from companies rather than new laws or federal rules. That contrasts with the Biden administration, which issued an executive order in 2023 requiring AI developers to share safety test results with the government before releasing powerful models.
Trump rescinded that order shortly after taking office, arguing it hindered innovation. The companies represented at Tuesday's meeting have each published their own safety frameworks in recent years, though the standards vary and enforcement is largely self-policed.
Nvidia, whose chips power most of the world's leading AI models, has seen its market value surge past $3 trillion. OpenAI, Anthropic and Google are racing to release more capable systems, and Meta has pushed open-source models that anyone can download and modify.
Trump did not announce new legislation, funding or a federal AI agency. He said the executives present were the right people to manage the technology's risks.
The White House has not released a full list of attendees or the text of the agreement beyond what Trump shared online. It is unclear whether the companies will face any consequences for failing to follow through.
