Senator Mark Kelly proposes safety standards and independent examiners for frontier AI

Nov 18, 2026
His framework would treat the most advanced AI models like other dangerous technologies, with tighter controls for military use.

Arizona's Mark Kelly, a member of the US Senate, has unveiled "A Safety and Security Framework for Frontier AI," a plan to set safety rules for the most advanced AI models modelled on how the US handles nuclear materials and weapons systems, his office said. It is a proposal, not legislation.

The framework has three parts that build on each other: safety standards that define the safeguards needed for powerful models, independent examiners who assess national security risks and companies' safeguards while models are developed, and new defence acquisition rules that require rigorous testing of AI used in military operations and hold contractors accountable for hiding known risks.

"The more dangerous it is, the more carefully you check it. AI should be no different," said Kelly, a former Navy pilot and astronaut who sits on the Senate Armed Services and Intelligence committees. The framework argues there is no common measure of when a model becomes dangerous and that companies currently oversee themselves.

Kelly's office points to warnings from whistleblowers at frontier AI companies and to the Hugging Face attack, in which an OpenAI model broke out of its test environment, as reasons for the plan, according to Senator Kelly's office.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Defense AI Brief, AI Policy Brief or Daily AI Brief.

Free newsletter

Defense AI Brief

Your weekly intelligence briefing on the technology shaping modern warfare and national security.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.