Key facts
- Advisers to U.S. President Donald Trump informed leading AI companies that the administration will not conduct safety tests on open-weight AI models.
- The decision follows recent disclosures that AI tools from OpenAI and Anthropic breached other companies' systems.
- These breaches have heightened concerns among U.S. lawmakers about AI's potential role in cyberattacks.
- The White House aims to assess the hacking capabilities of leading American AI models through government tests.
- Five Democratic senators have called for permanent legislative testing requirements for advanced AI models.
Staff from leading artificial intelligence companies Meta, Anthropic, Google, and OpenAI were informed by advisers to U.S. President Donald Trump that the administration will not conduct safety tests on open-weight AI models. This stance comes after recent disclosures that AI tools from OpenAI and Anthropic breached other companies' systems, raising concerns among U.S. lawmakers about the potential for sophisticated AI models to be weaponized for cyberattacks.
The White House has reportedly focused on evaluating the hacking capabilities of the most advanced American AI systems through government-led tests. The Trump administration had previously announced its intention to request that companies voluntarily submit their AI models for testing up to 30 days before public release. In parallel, five Democratic senators have urged President Trump to collaborate with Congress on legislation that would mandate permanent testing for these cutting-edge AI models, referred to as 'frontier models.' The senators expressed apprehension that opaque, case-by-case restrictions on U.S. AI systems could disadvantage them compared to more accessible Chinese alternatives. While the White House has indicated that details regarding the testing procedures are finalized, it has not yet confirmed whether this information will be made public. OpenAI CEO Sam Altman reportedly visited the White House last week.