Anthropic safety claims disputed after Pentagon stops using its AI

By Splitfeed EditorialLast updated Oct 7, 2026, 17:01 UTC

On October 7, 2026, the Pentagon ended its use of Anthropic's AI, with officials citing the company's safety guardrails as the reason for the decision. Around the same time, a firm called ClawSecure published test results that directly contradicted Anthropic's own published safety claims. The two developments arrived together, and Anthropic and the Department of Defense are now in an active dispute over the findings. A separate analysis flagged the episode as evidence of broader gaps in how the Pentagon evaluates and monitors the AI vendors it works with.

The collision between a government contract termination and an independent security firm's findings is what gives this story weight beyond a routine vendor dispute. Anthropic has built much of its public identity around safety research and responsible deployment, so a direct contradiction of its safety claims by an outside tester strikes at the core of that positioning. For the Pentagon, the episode raises questions about what due diligence was performed before the contract began and what oversight mechanisms exist to catch problems mid-deployment. If the gaps identified are systemic rather than specific to this relationship, the implications extend to other AI vendors working with military agencies.

No published reactions from the Pro-AI, Anti-AI or Middle Ground camps were available at the time of writing. The Pro-AI camp would typically argue in a case like this that independent testing methodologies may not reflect real-world deployment conditions and that contract disputes are a normal part of integrating new technology into large institutions. The Anti-AI camp would likely treat ClawSecure's findings as confirmation that AI safety claims from developers cannot be taken at face value and that military use of AI carries risks that outpace current governance. The Middle Ground camp would ordinarily call for clearer federal standards for testing and certifying AI systems before government adoption, treating the oversight gaps as the central problem to fix.

The most consequential next steps will be whether Anthropic or the Department of Defense releases a formal response to ClawSecure's methodology, whether congressional oversight bodies open inquiries into Pentagon AI vendor practices, and whether ClawSecure publishes more detailed findings that would allow independent review of its conclusions.