Trump Administration Reverses on AI Safety Testing After Anthropic Flags Mythos Risk

The Trump administration has reversed its stance on mandatory AI safety testing, reversing course just weeks after scrapping Biden-era safety protocols. The shift came after Anthropic declined to release its latest Claude Mythos model, citing concerns that advanced cybersecurity capabilities could be weaponized by malicious actors.

Previously, the administration had dismissed voluntary safety checks as regulatory overreach. In February, President Trump rebranded the US AI Safety Institute to the Center for AI Safety and Innovation (CAISI), removing "safety" from its name in what officials described as a push to accelerate AI development without government interference. However, White House National Economic Council Director Kevin Hassett indicated the administration now considers pre-release testing essential for frontier models that pose national security concerns.

Anthropic's decision to withhold Claude Mythos appears to have prompted internal reassessment. The company cited specific cybersecurity capabilities that, if released, could lower barriers for state-sponsored hackers and criminal enterprises. Hassett suggested Trump may issue an executive order requiring government review of advanced AI systems before public deployment, a position that directly contradicts the administration's earlier deregulatory posture. Industry observers note the reversal raises questions about whether the administration is responding to genuine security concerns or managing political fallout from the appearance of abandoning oversight entirely.