US group urges mandatory AI safety reviews before public release

A U.S. advocacy group is pressing the Trump administration to make frontier AI companies clear safety reviews before public release, with one crucial enforcement tool in mind: federal contracts. Americans for Responsible Innovation said the government should test cutting-edge models for security risks before deployment and deny lucrative public work to developers whose systems do not pass.

The proposal would put the Commerce Department’s Center for AI Standards and Innovation at the center of a mandatory review regime. It would cover firms spending at least $100 million a year on compute for frontier models or taking in at least $500 million a year from AI products and services, a threshold that reaches the biggest labs rather than the broader industry. In practice, that would make government procurement a gatekeeper for market access, forcing top AI developers to show their systems are safe enough for federal use before they can compete for contracts.

That is a more aggressive step than the current mix of voluntary testing arrangements. On May 5, 2026, CAISI announced agreements with Google DeepMind, Microsoft and xAI to give the U.S. government early access to frontier models for pre-deployment evaluation. Reuters reported that earlier partnerships with OpenAI and Anthropic dated to 2024 and were renegotiated under the Trump administration. CAISI has said its evaluations and targeted research are meant to better assess frontier capabilities and advance AI security, and The Hill reported the office had already completed more than 40 evaluations.

The policy debate comes as officials weigh the risks of models that can make cyberattacks faster and easier to execute. Reuters reported that the government was particularly alarmed by Anthropic’s Mythos because of its potential cyber capabilities, while POLITICO reported that White House officials had been meeting privately with tech and cyber firms about those concerns. That suggests the administration is already moving toward a more hands-on posture, even before any formal rulemaking.

The wider significance is political as much as technical. The 2025 AI Action Plan directed CAISI toward a central role in the nation’s evaluation system, building on a framework that began as the U.S. Artificial Intelligence Safety Institute under the Biden administration. A mandatory review system would not create AI testing from scratch. It would harden existing voluntary checks into a procurement standard, potentially setting a de facto national benchmark for the largest model developers and giving Washington its first meaningful leverage over how frontier systems are cleared for release.