The Trump administration will ask leading AI developers to voluntarily submit their most capable models for government cybersecurity tests before releasing them to the public, according to an executive order released Tuesday, as security fears mount over powerful new AI systems such as Anthropic’s Mythos.
Trump signed an executive order directing the departments of Treasury, Defence, Commerce and Homeland Security, plus other government officials and agencies, to secure agreements with AI developers to test their models.
Agencies would get up to 30 days to test models before release to outside organizations and are directed to bolster cyber defence across government.
Shift In Trump’s AI Strategy
The order signals Trump shifting his strategy on AI and taking a more active role in monitoring the technology’s capabilities.
Since returning to office, he has favoured a hands-off approach to the tech sector and tried to discourage states from adopting AI regulations he opposes.
The decision to implement voluntary testing could hurt industry profits if it slows the rollout of new models or prompts companies to change how they perform to address security concerns.
Anthropic, OpenAI and Alphabet’s Google met with the U.S. government about cybersecurity, a senior official said in May.
Google’s Kent Walker called the order “an important step forward,” while Anthropic said it looked forward to supporting its implementation. OpenAI CEO Sam Altman said it “gets the balance right,” adding: “The US should lead on AI by continuing to develop the very best models, making sure they’re safe, and getting cyber tools into the hands of trusted defenders.”
Trump was slated to sign the order on May 21 but postponed, saying he did not like certain aspects and did not want to undermine the U.S. position in its AI competition with China.
Protecting Vital U.S. Sectors
Treasury Secretary Scott Bessent, who consulted with banks during the order’s development, was directed to work with AI developers and critical infrastructure providers to scan software for cybersecurity flaws and develop patches. Critical infrastructure refers to sectors vital to the U.S. economy deserving special protection, such as banks, emergency services and hospitals.
Voluntary federal testing has been in place for a few years, with OpenAI and Anthropic submitting their models to the Commerce Department’s Centre for AI Standards and Innovation.
In May, Google, xAI and Microsoft also agreed to submit their models for security testing, though the details later disappeared from the department’s website.
(with input from Reuters)
