Senator Mark Kelly proposes a frontier AI framework with hazard levels, embedded federal examiners and defense AI liability

Office of Sen. Mark Kelly · Policy report

Read the source (opens in a new tab)
What happened
Senator Mark Kelly published "A Safety and Security Framework for Frontier AI" on November 17, 2026. It is a policy proposal, not a bill.
Key points
  • NIST, the national laboratories and security and health agencies would set AI hazard levels covering weapons of mass destruction, cyberattacks and loss of control, published within 6 months.
  • Independent federal examiners would be embedded at developers of the most advanced models, with full access to facilities and models, and a remit limited to national security.
  • AI for high-risk military and intelligence missions would need documented risk assessments and independent testing; contractors that knowingly conceal serious risks could face civil liability for resulting harm.
  • Kelly writes that models "that could help cure deadly diseases can also help someone develop biological weapons or launch devastating cyberattacks," while companies decide for themselves which are dangerous.
Evidence
A senator's own policy framework, modeled on how the US oversees nuclear materials, dangerous pathogens and weapons systems. It states goals and mechanisms, not legislative text.
Uncertain
The framework and its press release announce no bill, cosponsors or timeline.

See it in the latest record