Hassett Says White House Looking into AI Security Executive Order

The White House is studying an executive order on artificial intelligence safety, with the goal of establishing a clearer pre-release review pathway for new AI models. Kevin Hassett, the Chairman of the National Economic Council, stated on May 6th that AI models with potential future cybersecurity risks may need to undergo testing and prove their safety before they can be made available for general use.

The direct catalyst for this round of policy discussions is Anthropic's Mythos model. This model excels at discovering cybersecurity vulnerabilities but could also be used to create broader cybersecurity risks, which is why Anthropic currently grants access only to a select few large tech and financial companies.

Mythos has pushed the boundaries of AI regulation in the United States to new limits. Hassett mentioned that the government has coordinated with the public and private sectors to test the model, focusing on confirming it will not harm American businesses and government networks before opening access on a large scale.

“We are considering possibly through an executive order, providing everyone with a clear roadmap.”

If the executive order is implemented, the impact could extend beyond just Anthropic. Hassett indicated that Mythos might only be the first case, and the testing mechanism "very likely" will eventually be expanded to all AI companies.

If this mechanism is put into place, the core variables for AI companies will no longer be solely the model's capabilities and commercialization speed, but also include government testing, compliance processes, and cooperation terms with federal agencies. The U.S. Department of Commerce has expanded its voluntary testing program on May 5th, with Google, Microsoft, and xAI agreeing to open their models to the government for capability assessment and security improvements; OpenAI and Anthropic have previously joined this project.