Anthropic: No Kill Switch for Pentagon-Deployed AI
SkimNews Take
The Pentagon's "supply-chain risk" designation for Anthropic's AI highlights a broader challenge in integrating advanced, self-contained AI systems into critical infrastructure: the difficulty of maintaining human oversight and control over autonomous operations once they are deployed.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic told a federal appeals court in D.C. it has no visibility, technical ability, or 'kill switch' to control or shut down its AI models once the Pentagon deploys them, and noted the Pentagon has the opportunity to test models before deployment.
- The Pentagon designated Anthropic a 'supply chain risk,' arguing the AI firm is inappropriately inserting itself into how its technology is used in sensitive military operations.
- Anthropic's usage policies bar Claude from autonomous weapons or mass surveillance — red lines the Pentagon dismissed as 'red herrings,' sparking the dispute.
- A D.C. court rejected Anthropic's bid to pause the supply chain risk label, while a California judge granted a similar request in a parallel case, leaving Anthropic barred from new Pentagon contracts but free to work with other federal agencies.
- The Trump administration is racing to deploy the new 'Mythos' model across government, and now agency heads are scrambling to protect their systems from cyber attacks using Mythos — a dynamic that could undercut the Pentagon's national security argument against Anthropic.
- A hearing is scheduled for May 19.
Why it matters: The filing flips the Pentagon's framing: if Anthropic truly cannot touch models once deployed, the 'supply chain risk' label rests on its policy stances — bans on autonomous weapons and mass surveillance — rather than any technical control. The administration's own cyber-scramble to secure agencies against the Mythos model it champions reinforces that tension heading into the May 19 hearing.
