Pentagon CIO Forces Anthropic Purge as Defense AI Dispute Exposes Regulatory Vacuum
A career bureaucrat's 180-day removal order crystallizes internal Pentagon tensions over frontier AI access, while Congress remains sidelined in a conflict reshaping defense-industrial policy.
Pentagon Chief Information Officer Kirsten Davies has ordered the removal of Anthropic’s AI systems from all military networks within 180 days, escalating a standoff that reveals deepening fractures between defense leadership and a frontier AI firm over surveillance and autonomous weapons—and exposing the absence of congressional guardrails for military AI deployment.
The CBS News-obtained March 6 memo, signed by Davies one day after Defense Secretary Pete Hegseth formally designated Anthropic a supply chain risk, warns that adversaries “can exploit vulnerabilities” posing “potential catastrophic risks to the warfighter.” The directive mandates removal from nuclear weapons, ballistic missile defense, and cyber warfare systems—and bars all Pentagon contractors from using Claude on defense-related work by September 2, 2026.
The dispute centers on Anthropic’s restrictions prohibiting Claude from powering autonomous weapons or mass Surveillance—red lines the Pentagon says are incompatible with its requirement that AI vendors allow “any lawful use.” According to CNBC, Anthropic was awarded a $200 million contract in July 2025 and became the first AI company deployed on classified networks, partnering with Palantir to embed Claude across sensitive defense workflows. Yet the Pentagon’s reliance on a single vendor created what Under Secretary Emil Michael described as a “whoa moment” for leadership when they realized operational dependency during active combat in Iran.
Career vs. Political: Davies Emerges as Enforcer
Davies, a career Pentagon information officer, now controls the chokepoint. According to The Hill, Davies is the sole official authorized to grant exemptions, which will only be considered for mission-critical activities with no viable alternative and require comprehensive risk mitigation plans. The narrow exemption pathway signals a policy of default removal rather than accommodation—a harder line than Trump’s six-month federal phaseout.
While Hegseth delivered the public ultimatum, Pentagon CTO Emil Michael drove negotiations, spending three months with Anthropic compared to one or two weeks with competitors, according to CBS News. Michael, a former Uber executive appointed as Under Secretary for Research and Engineering, publicly called Anthropic CEO Dario Amodei a “liar” with a “God-complex” on X as talks collapsed. By March 12, Michael told CNBC there was “no chance” of renewed negotiations, citing “leaking and bad faith negotiation”.
The dynamic illustrates career-political appointee tension: Davies wields statutory authority over information systems, while Michael—a political hire with venture capital ties—shapes AI acquisition strategy. Davies’s memo operationalizes Hegseth’s broader directive, but her centralized exemption authority suggests bureaucratic gatekeeping could outlast political turnover.
OpenAI’s Replacement Bargain Under Scrutiny
Hours after Trump’s ban, OpenAI announced a Pentagon deal claiming identical red lines on surveillance and autonomous weapons. According to MIT Technology Review, OpenAI CEO Sam Altman said the difference was approach: “Anthropic seemed more focused on specific prohibitions in the contract, rather than citing applicable laws”—a distinction legal experts view as substantively weaker.
George Washington University Law School’s Jessica Tillipman noted OpenAI’s contract “does not give OpenAI an Anthropic-style, free-standing right to prohibit otherwise-lawful government use”. OpenAI’s agreement references existing laws but leaves interpretation to the Pentagon, which controls both enforcement and legal interpretation. Altman later amended the contract to clarify that “the AI system shall not be intentionally used for domestic surveillance” and that NSA is excluded from this specific contract, but Electronic Frontier Foundation flagged “weasel words” like “intentionally” and “unconstrained” as loopholes.
The rapid OpenAI pivot prompted internal dissent: one OpenAI employee told CNN many staff “really respect” Anthropic for standing firm and are frustrated with their own leadership’s handling. Robotics lead Caitlin Kalinowski resigned March 7, citing concerns over “surveillance of Americans without judicial oversight and lethal autonomy without human authorization”.
| Company | Classified Access | Contract Stance | Status |
|---|---|---|---|
| Anthropic | First approved (July 2025) | Explicit prohibitions on surveillance, autonomous weapons | Supply chain risk; 180-day removal |
| OpenAI | Approved Feb 2026 | “Any lawful use” + reference to existing laws | Active contract |
| xAI (Musk) | Approved Feb 2026 | “Any lawful use” standard | Active; expanding integration |
| Under review | Negotiating terms | Seeking classified clearance |
Congress Sidelined Despite Bipartisan Unease
The dispute unfolded with minimal legislative input, despite jurisdiction over both defense authorization and surveillance law. According to Axios, Senate Armed Services member Thom Tillis called the public confrontation “sophomoric” and questioned “Why isn’t this occurring in a boardroom?” Tillis added “it’s fair to say Congress needs to weigh in if they have a tool that could actually result in mass surveillance”.
Democratic senators went further. Sen. Mark Kelly warned “DOD is trying to strong-arm Anthropic into providing every tool they have to surveil U.S. citizens… That’s unconstitutional. That’s not the role of the Department of Defense”. A March 5 letter from former defense officials to Congress, reported by CNBC, called for investigation, arguing “Blacklisting one of America’s leading AI companies—and requiring its thousands of contractors to sever ties—does not strengthen our competitive position. It weakens it”.
Yet no hearings have been scheduled. The FY2026 NDAA directs the Pentagon to establish an AI model assessment team by June 2026 and develop a department-wide framework by June 2027, but provides no interim standards for procurement disputes. Lawfare analysts argue “this fight is happening because Congress hasn’t set substantive rules for military AI. If Congress had legislated guidelines on autonomous weapons and surveillance, Anthropic would likely be far more comfortable selling its systems to the military”.
The supply chain risk designation—historically reserved for foreign adversaries like Huawei—has never before targeted a U.S. company. Hegseth’s move came immediately after Trump’s Truth Social directive ordering “EVERY Federal Agency” to “IMMEDIATELY CEASE” using Anthropic. Legal experts note the designation lacks procedural foundation: statute requires consultation, written determination with three mandatory findings, and interagency coordination—none of which appear to have occurred. The expedited process suggests political override of administrative procedure.
Anthropic’s Billion-Dollar Legal Gambit
Anthropic filed a 48-page lawsuit March 9 in Northern District of California, arguing the government’s actions are “unprecedented and unlawful” and violate First Amendment protections. According to The Hill, more than 100 enterprise customers contacted Anthropic with concerns, and CFO estimates put 2026 revenue impact in the hundreds of millions to billions.
The secondary boycott effect is severe: if interpreted strictly, any company doing any business with DOD cannot use Claude—potentially forcing Amazon, Google, and Nvidia (Anthropic investors with Pentagon contracts) to divest. Microsoft filed an amicus brief March 11 supporting Anthropic’s legal position.