U.S. officials have signed a framework to vet new AI models from DeepMind, Microsoft, and xAI before public launch. The Commerce Department’s Center for AI Standards and Innovation will run pre‑deployment tests and targeted research, aiming to gauge how cutting‑edge systems might affect national security. These reviews will cover cyber, bio, and chemical weapon risks.

Earlier scrutiny stemmed from Anthropic’s Mythos model, which reportedly outperforms rivals in detecting cyber‑security flaws. The start‑up has faced a national‑security label after refusing Pentagon access and is suing the administration over the designation. A recent meeting between Anthropic’s chief Dario Amodei and White House staff signaled a thaw. The pact mirrors a 2022 agreement that granted early model access for safety reviews.

Under the new deal, the agency will receive models with reduced safeguards to test capabilities that could threaten cyber, bio, or chemical security. Findings will be shared with the Pentagon, White House, and intelligence agencies. The initiative aims to scale rigorous assessment science, the director said, noting it protects investors by clarifying risk exposure before products hit markets.

These reviews could reshape market dynamics, as firms must navigate additional compliance layers before launching AI tools. Investors watching the tech sector will note that early access and safety assurance could become a competitive advantage. The U.S. move signals a broader push to align commercial AI growth with national defense priorities.