Technology & Science
Pentagon Gives Anthropic Friday Ultimatum, Floats Defense Production Act
On 24 Feb 2026, Defense Secretary Pete Hegseth told Anthropic it must drop its self-imposed bans on autonomous targeting and U.S. surveillance by 27 Feb or face contract cancellation, blacklisting, and possible seizure of its Claude AI under the Defense Production Act.
Focusing Facts
- Anthropic’s $200 million contract, awarded summer 2025 for Claude’s use on classified networks, is explicitly on the chopping block.
- Claude is currently the only frontier-scale commercial AI model deployed inside the Pentagon’s classified systems; rivals Google, OpenAI and xAI remain in unclassified environments.
- Hegseth threatened to label Anthropic a “supply-chain risk,” which would bar it and its subcontractors from future defense work even beyond the disputed deal.
Context
Washington’s coercive posture evokes President Truman’s 1952 seizure of steel mills during the Korean War and the 2016 Apple–FBI encryption showdown: moments when the U.S. state asserted primacy over private tech seen as vital for national security. The clash sits at the intersection of two accelerating trends—the militarization of commercially developed AI and the century-long cycle in which cutting-edge dual-use technologies (radio in WWI, nuclear physics in the 1940s, cyberspace in the 1990s) migrate from civilian labs to the battlespace faster than legal or ethical frameworks mature. Whether the Pentagon or a safety-focused vendor sets the guardrails will shape precedent for autonomous systems throughout NATO and, by extension, China’s and Russia’s doctrines. If the government compels access now, future founders may relocate or bake secrecy into models, while a concession by Anthropic could normalize unrestricted state control of frontier AI—either outcome echoing through the next 100 years of civil-military tech relations.
Perspectives
Right-leaning US media
e.g., Fox News — Coverage stresses that the Pentagon simply wants unfettered, lawful access to Claude for national-security missions and depicts Anthropic’s limits as an obstacle the military cannot tolerate. By echoing Republican talking points about eliminating “woke” constraints and foregrounding Hegseth’s comparison to aircraft usage, the reporting sidelines civil-liberties worries and presents the stand-off as a matter of corporate insubordination rather than ethical debate.
Mainstream/liberal outlets
e.g., BBC, The Boston Globe — Stories underline Anthropic’s self-imposed red lines on autonomous weapons and mass surveillance and frame Hegseth’s threat to invoke the Defense Production Act as an aggressive bid to erode AI safety guardrails. By foregrounding Anthropic’s ‘responsible AI’ branding and quoting critics who say the law lags behind technology, these pieces risk amplifying the company’s PR narrative while giving less space to the Pentagon’s claim that it always follows legal use policies.
International business press
e.g., The Economist, Australian Financial Review — Reporting casts the confrontation as a high-stakes commercial power struggle, noting that Anthropic has unusual leverage despite Hegseth’s ultimatum and that wartime powers like the Defense Production Act could reshape tech procurement. The market-oriented lens tends to reduce the moral dimension to contract risk and leverage calculations, implicitly treating the militarisation of AI as an inevitability driven by competitive advantage.
Like what you're reading?