POLISCOPE
Back to feed
NewsThe HillMay 6, 2026United States

White House Considers FDA-Style Safety Reviews for AI Models

The Trump administration is considering an executive order to mandate safety evaluations for advanced AI models, similar to the FDA's drug approval process. This potential shift in policy follows security concerns regarding Anthropic's new 'Mythos' AI model, which can identify significant software vulnerabilities.

Artificial IntelligenceCybersecurityExecutive OrderRegulationTechnology Policy
Read the full story at The Hill

Why It Matters

If implemented, this policy would require AI developers to prove their models are safe before public release, potentially changing how AI software is developed and deployed in the United States.

Key Facts

  • The White House is considering an executive order to require AI models to undergo safety evaluations similar to FDA drug approval processes.
  • The proposal is a response to security concerns regarding Anthropic's 'Mythos' AI model, which can identify decades-old security vulnerabilities in software and infrastructure.
  • Anthropic released Mythos to a limited set of companies, including Wall Street banks, last month.
  • NIST has reached agreements to test AI models from xAI, Microsoft, and Google DeepMind, in addition to existing agreements with OpenAI and Anthropic.
  • Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell have convened financial leaders to discuss cybersecurity risks associated with the model.
  • The potential oversight plan represents a shift from the administration's previous 'light-touch' approach to AI innovation established in 2024.
  • Mythos is currently being tested and used by some federal agencies.
  • The administration is coordinating an 'all-of-government' effort to ensure the model is tested before any broader public release.

Who's Mentioned

organizationAnthropic“Developer of the Mythos AI model”personScott Bessent“Treasury Secretary”organizationNIST“National Institute of Standards and Technology”otherMythos“Advanced AI model capable of identifying security vulnerabilities”personKevin Hassett“National Economic Council Director”organizationCISA“Cybersecurity and Infrastructure Security Agency”date2026“Copyright year”personJerome Powell“Federal Reserve Chair”