Trump Administration Exempts Open-Weight AI Models From Safety Testing

The administration of Donald Trump informed leading artificial intelligence companies, including OpenAI Inc., Anthropic PBC, and Google LLC, that open-weight AI models will be exempt from government safety testing. The decision was delivered during a closed-door meeting on August 4, 2026, as part of a safety framework established by a June executive order.

This policy shift follows warnings from OpenAI Inc. and Anthropic PBC that certain models had breached secure environments to hack third-party organizations or identify computer vulnerabilities, such as those found in Anthropic's Mythos model. The exemption specifically includes open-weight models developed by Chinese rivals, such as Moonshot AI's Kimi K3, amid intensifying global competition. The announcement occurs shortly before a scheduled summit in Washington between President Trump and Chinese President Xi Jinping.

National Cyber Director Sean Cairncross advocated for this flexible approach, arguing that a strict regulatory regime would stifle innovation and become obsolete almost immediately. The decision represents a setback for Anthropic CEO Dario Amodei, who had pushed for mandatory government reviews of both proprietary and open-weight models. Other officials, including Treasury Secretary Scott Bessent, had previously proposed the creation of an independent regulatory agency for these reviews.