DoD-Anthropic Dispute Threatens AI Nuclear Safety Research, Report Says

The U.S. Department of Defense's decision to label Anthropic a "supply chain risk" is now threatening critical AI safety research related to nuclear weapons, according to Fast Company.

The report, published recently, notes that projects at the Department of Energy designed to limit AI's work on nuclear weapons may suddenly halt as government agencies struggle to understand whether they're allowed to use Claude.

Anthropic had partnered with the DOE's National Nuclear Security Administration (NNSA) last April to evaluate its AI models for nuclear proliferation risks. According to Anthropic, the purpose of the work was to "evaluate our AI models for potential nuclear and radiological risks."

The concern: developing nuclear weapons requires specialized knowledge, and as AI models become more capable, there's a risk they could provide users with dangerous technical information.

The DOE partnership was designed to assess whether Claude could be used to help build nuclear weapons—and to develop safeguards against such misuse. Now, with the DoD designation, that work is in limbo.

The DoD designation came after weeks of tense negotiations between the Pentagon and Anthropic over how the military could use the company's AI. Anthropic has filed a lawsuit challenging the designation, arguing it's "legally unsound" and exceeds the Secretary's authority.