The Trump administration told AI developers on Tuesday that it will not put open-weight AI models through voluntary safety tests, according to two sources familiar with the discussions.
Open models, including Nvidia’s Nemotron and Meta’s Llama, are AI systems with publicly accessible core components. Closed models are controlled by specific companies. Major U.S. developers of closed models are OpenAI, Google and Anthropic.
The White House discussed the unpublished testing rules with staff from Meta (META.O), Anthropic, Google (GOOGL.O), Nvidia and Open AI, according to five sources familiar with the meeting.
The meeting follows disclosures from Open AI and Anthropic that their AI tools breached the systems of other companies. The hacks raised concerns among U.S. lawmakers about whether increasingly capable AI models could be used to conduct or facilitate cyberattacks. Open AI on Tuesday said the company’s models were involved in another cybersecurity incident during testing and that the situation involved AI systems accessing the internet.
Sharing the framework with a small group of companies without making it public “deepens a serious gap in federal oversight, leaving open questions about how the federal government plans to evaluate the safety and security of advanced AI models,” said tech policy advocacy group Americans for Responsible Innovation. The group pushes U.S. officials to implement policies that “protect the public from harm,” according to its website.
The Trump administration said in June the tests would be voluntary, and designed for AI models with sophisticated hacking capabilities. Open AI and Anthropic both have models with advanced cyber security capabilities.
Five Democratic senators called on Trump on Tuesday to work with Congress to pass legislation making testing permanent for the most advanced American-made AI models, called “frontier models.”
“The United States cannot afford to create a policy environment in which the most advanced American AI systems are subject to opaque, case-by-case restrictions while Chinese alternatives appear cheaper, easier to access, and more predictable to deploy,” the senators said in the letter.
The Trump administration has said little in public beyond that U.S. officials are monitoring the Open AI hack. Open AI CEO Sam Altman visited the White House last week. A White House official said on Monday the Trump administration has finalized the details of voluntary cybersecurity tests to measure the hacking capabilities of the most advanced American AI models, and is planning to discuss them with the AI industry. The official did not indicate who would attend the discussions.
Meta was invited, a company spokesperson said, as were Anthropic and Open AI, according to two sources familiar with the meeting.
The White House did not provide details about the tests, including how results would be reported, the metrics used and whether any of it would be made public.
A group of 15 Republican state attorneys general on Monday asked Open AI to preserve all potentially relevant documents related to its disclosure that its AI system escaped containment and hacked AI company Hugging Face. Citing a Reuters report that the rogue agent in one case left notes for how future versions of itself could escape internal guardrails, they wrote that the company may have violated state consumer protection laws.
OpenAI said in a statement it takes the letter from the attorneys general seriously and will share a technical report about the Hugging Face attack after it completes a review.
The U.S. House of Representatives’ cybersecurity committee on Monday asked OpenAI’s Sam Altman to brief them on the attack on Hugging Face.
Altman visited the White House last week to discuss details of the voluntary tests and the company’s upcoming AI products, the company said in a statement. In a separate statement on Monday, the company said it had asked the Trump administration to put the Commerce Department’s AI safety specialists at the center of any cybersecurity testing. The company pointed to China, whose government has a more centralized strategy on AI compared with the U.S.
The Information, a tech publication, reported on Monday that the Trump administration invited representatives from Google. A Google spokesperson declined to comment.
U.S. President Donald Trump directed his team in June to write a series of tests to assess the hacking capabilities of the most advanced American AI systems.
The Trump administration has had a rocky relationship with Anthropic. The company earlier this year refused to allow the U.S. military to use its AI models for domestic surveillance and fully autonomous weapons systems, and the government retaliated by putting it on a national security blacklist.
Anthropic said last week that some of its AI models hacked into the systems of three companies during cybersecurity tests. That disclosure followed rival OpenAI’s report that one of its AI agents escaped a testing environment and hacked into the systems of the AI company Hugging Face.
