Pentagon Drops Anthropic Claude Over Safety Rules
Key insights
- Hegseth declared Anthropic a supply-chain risk in March, three days after it refused to remove guardrails on weapons and surveillance.
- The Pentagon is testing OpenAI, Google, and xAI's Grok as replacements, evaluated by 25 departmental power users across classified systems.
- Anthropic is contesting the supply-chain-risk designation in court, arguing it could cost the company billions in government revenue.
Why this matters
Claude was not a peripheral tool but the primary AI underpinning the Maven Smart System, the Pentagon's classified operations platform, and its removal signals that safety-first AI labs face structural disadvantage when competing for the most powerful government buyers. The supply-chain-risk designation creates a legal and commercial precedent: governments can formally sideline AI vendors whose guardrails conflict with operational requirements, potentially reshaping how labs position products for defense contracts. Anthropic's court challenge, where billions in government revenue are at stake, will determine whether safety commitments can survive large-scale military procurement and set a template for the rest of the frontier AI industry.
Summary
Anthropic's Claude was the Pentagon's primary AI on classified networks, built into the Maven Smart System for sensitive operations. Defense Secretary Pete Hegseth declared Anthropic a supply-chain risk in March after the company refused to remove guardrails against mass surveillance of Americans and autonomous weapons targeting.
Essentially: (OpenAI, Google, xAI Grok) are now being evaluated as replacements by 25 Pentagon power users.
- Claude underpinned Maven, reportedly used for classified operations including recent strikes
- Anthropic is challenging the designation in court, arguing it could cost billions in revenue
The Pentagon's choice tests whether a safety-first AI brand can survive the demands of a major military customer.
Potential risks and opportunities
Risks
- Anthropic loses its existing Pentagon contract revenue if the supply-chain-risk designation survives the court challenge, with billions at stake per Anthropic's own legal filing
- Maven Smart System operations face classified disruption during transition if no replacement model achieves equivalent integration depth before the DoD requires it operationally
- OpenAI or Google, by accepting Pentagon contracts on guardrail terms Anthropic refused, risk internal researcher departures and reputational damage in enterprise markets where safety posture matters
Opportunities
- OpenAI, Google, and xAI Grok stand to inherit Anthropic's primary position in Pentagon classified AI workloads if their models clear the 25-power-user evaluation
- Defense AI systems integrators working with Maven Smart System gain leverage as migration architects during a complex classified-platform transition
- Anthropic can use the litigation and public framing to strengthen its safety brand in enterprise and consumer markets where AI limits matter to buyers
What we don't know yet
- Which model, if any, passes the 25-power-user evaluation and on what timeline, given Claude's deep Maven integration could take months to undo
- Whether Anthropic's lawsuit has a realistic path to reversing Hegseth's supply-chain-risk designation before a full replacement is deployed
- What 'recent strikes' means specifically in terms of Claude's role: active targeting support, or logistics and intelligence analysis
Originally reported by techtimes.com
Read the original article →Original headline: Pentagon Racing to Replace Anthropic's Claude in Classified Military Systems — 'Too Safe for War'