WASHINGTON -- The Trump administration's new guidelines for powerful artificial-intelligence tools exempt open models made by U.S. companies, a gap that marks the White House's latest bid to ensure AI tools are safe without slowing down innovation.
Under a framework that administration officials discussed with executives from leading AI companies Tuesday, only makers of closed, proprietary U.S. models that demonstrate state-of-the-art capabilities in cybersecurity and hacking based on performance benchmarks would have to voluntarily submit those models to the government for testing before they are released, people familiar with the matter said.
Open-weight models, whose developers, including Nvidia, make them available to download so users can run and customize them on their own, would be exempt.
The general definition of state-of-the-art capabilities could be interpreted differently by makers of closed models and the White House, potentially creating confusion about which models fall into the category, the people said. Tools from Anthropic, OpenAI and Alphabet's Google that are among the most powerful available are likely to force those companies to work with the administration, while other companies, such as Elon Musk's SpaceX and Facebook parent Meta Platforms might not have to, they said.
The prerelease testing is a core part of a June executive order that marked President Trump's attempt to address the mounting security risks posed by AI. While the testing is voluntary, industry executives have said it would be risky to bypass the government review, particularly if there were safety concerns raised after the fact.
OpenAI recently delayed its latest class of tools while the administration completed the framework, while Anthropic had two versions of its Mythos model shut down after the White House raised concerns about their security.
Nvidia and other makers of open-weight models initially will be exempt but might have to eventually submit their tools for testing as they get more powerful, the people said. Nvidia CEO Jensen Huang, a staunch proponent of open-weight tools, has been among the executives discussing AI oversight with the White House in recent weeks. Nvidia designs chips needed to run models and makes AI tools.
"Companies that choose to collaborate with the administration through this framework are putting American innovation, security and cyber defense first," White House spokeswoman Liz Huston said.
OpenAI and Anthropic have said that their models went rogue during training in recent months and hacked other companies, highlighting the rapid advancement in the latest tools.
Impressive new open-weight tools from Chinese companies that are cheaper to use have also raised the stakes for AI regulation. The administration is considering restrictions on some of the Chinese companies, which are accused of training their models using recent advances by U.S. competitors in an unfair way. Some executives at Anthropic and OpenAI have pushed for the crackdown.
But Huang and many other tech leaders have stressed the importance of open-weight models, with Trump caught in the middle, hoping AI and the data centers that drive them can continue benefiting the economy and elevating the U.S. in the tech race.
"China has virtually no controls," Trump said in the Oval Office last week. "It's freewheeling a little bit. So we have to be careful in both ways. We don't want to restrict them where all of a sudden we come in second to China."
Axios earlier reported on the open-model exemption.
National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent are among the officials leading the AI work and increasing attention on security risks, while White House AI adviser David Sacks, a venture capitalist, has advocated for a lighter touch that lets models naturally proliferate and be used to defend against cyberattacks and other threats.
Skeptics of the administration's approach say the voluntary pledges are inadequate to contain the risks posed by new models and that the plan could unfairly benefit the smaller companies that don't have to follow the guidelines.
Determining which models must follow stricter rules has proved challenging for years. The Biden administration used a quantitative approach to gauge the power of AI models, but critics said the measure was incomplete and lacked context.
Executives at the AI companies, including OpenAI CEO Sam Altman and Dario Amodei of Anthropic, say they welcome more government involvement in the industry as long as the system is fair and transparent. They have called for the administration to work with governments around the world to create global AI oversight that gives consumers more confidence in the industry. Many are skeptical of the technology, fueling a backlash in the U.S.
Comments