WASHINGTON –
President Donald Trump on Friday ordered every U.S. federal agency to cease using Anthropic’s AI, and Defense Secretary Pete Hegseth said he would designate the San Francisco company a “supply‑chain risk,” a label typically applied to foreign vendors. The move follows Anthropic’s refusal to let the Pentagon use its Claude AI system “for all lawful purposes,” including potential applications the company says cross its safety red lines. (trumpstruth.org)
“The Leftwing nut jobs at Anthropic have made a DISASTROUS MISTAKE trying to STRONG-ARM the Department of War,” Trump wrote on Truth Social, adding there will be a six‑month phase‑out for agencies where Anthropic is embedded. (trumpstruth.org)
Anthropic chief executive Dario Amodei responded that some uses of AI are “simply outside the bounds of what today’s technology can safely and reliably do,” reaffirming two non‑negotiables: no mass domestic surveillance and no fully autonomous weapons. Hegseth, in an X post, told defense contractors that “no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic,” while allowing up to six months to off‑board systems already in use. (anthropic.com)
Why this standoff resonates beyond Washington
The rupture pits a White House pursuing “AI‑first” military integration against a frontier‑model developer insisting on contractual limits-an argument with immediate consequences for U.S. and allied operations and for how far elected officials can lean on private firms to enable contested uses of AI. Claude has been integrated into the Pentagon’s Maven Smart System and, according to multiple accounts, supported planning for the January 3, 2026 raid that seized Venezuelan President Nicolás Maduro; reporting also indicates Claude aided recent U.S. targeting and operational analysis linked to strikes on Iran. NATO is concurrently fielding a Maven Smart System variant across Allied Command Operations. Replacing a deeply embedded model at scale in six months is a non‑trivial task for any command, raising questions about continuity of targeting support, intelligence workflows, and coalition interoperability.
An unprecedented label for a U.S. tech firm
Hegseth’s “supply‑chain risk” designation-common in past actions against Huawei or Kaspersky-is rare for an American company and could force primes and subcontractors to unwind ties with Anthropic or risk losing Pentagon work. Historically, such actions have moved through the Federal Acquisition Security Council or agency‑specific authorities (for example, directives that removed Kaspersky from federal networks), with formal findings, notice to Congress, and appeal mechanisms. Tech policy lawyers also note that a social‑media declaration is not, by itself, a completed exclusion order; contracting officers still have to translate it into specific instructions, and federal courts remain a backstop if Anthropic challenges the move as arbitrary or discriminatory.
The administration’s step caps a week of escalations in which officials signaled both the “supply‑chain risk” route and use of the Defense Production Act to compel access to Anthropic’s systems. Senator Edward Markey urged Congress to reverse the designation, and former NSA Director Paul Nakasone warned that blacklisting a domestic AI supplier risks long‑term damage to Pentagon-Silicon Valley trust at the very moment the U.S. is trying to secure preferred‑partner status with leading AI labs.
Inside the government’s ‘any lawful purpose’ demand
Defense leaders argue the military cannot accept vendor‑imposed vetoes on warfighting and that existing law already constrains activities like domestic surveillance and autonomous engagement. In their view, once Congress has authorized a mission and commanders are operating within U.S. and international law, private companies should not retain the contractual right to narrow how tools are used in theater.
Model‑use clauses built around “all lawful purposes” have now surfaced in OpenAI’s revised agreement, which the company says bars domestic surveillance and autonomous‑weapon use and keeps NSA access out of scope unless renegotiated. Procurement analysts note that restating legal boundaries in a contract can strengthen remedies if the government breaches those terms, though it does not grant a vendor an open‑ended right to block otherwise lawful use. The Anthropic dispute is therefore emerging as a test case for a broader question: how far major AI providers will go in embedding substantive “red lines” into standard government contracts, and whether future administrations will tolerate that constraint.
Amid the uproar, OpenAI chief executive Sam Altman told employees the company shares Anthropic’s “red lines,” and he publicly conceded its initial Pentagon deal looked “opportunistic and sloppy” before being amended to harden safeguards. In an internal note reported by Sky News, Altman wrote: “Regardless of how we got here, this is no longer just an issue between Anthropic and the DoW; this is an issue for the whole industry and it is important to clarify our stance.” The remarks underscore how quickly a bilateral contract fight has become a de facto industry‑wide debate over minimum safety baselines for national‑security work.
Where global rules and battlefield realities collide
- The U.S. military’s governing policy for autonomous weapons, DoD Directive 3000.09 (updated 2023), requires “appropriate levels of human judgment over the use of force,” but it does not categorically ban autonomous lethal engagement if rigorous approvals are met-one source of continued confusion inside and outside the Pentagon. The directive leaves room for systems that select and engage targets on their own under certain conditions, precisely the zone where Anthropic says current AI is too brittle to be trusted.
- In Europe, the EU’s AI Act entered into force on August 1, 2024, with staged application into 2026-2027; crucially, it exempts AI “used only for military or national security” from scope, limiting direct leverage over U.S. defense AI deployments even as it tightens rules on civilian and dual‑use systems.
- UN deliberations on lethal autonomous weapons continue under the Convention on Certain Conventional Weapons, where a majority of states back a binding instrument but negotiations have stalled, highlighting the gap between ethical consensus and enforceable law. That gap is now colliding with real procurement decisions in Washington, as officials seek flexibility that global norms have yet to crystallize or constrain.
Operational and industrial fallout
Palantir’s Maven Smart System-the data backbone where Claude is integrated-has active contracts through 2029 and expanding adoption across combatant commands and NATO. Any severing of Anthropic ties could force rapid system refits, model swaps, or narrowed use cases for mission planning, targeting support, and logistics simulations. Program officials now face a choice between accelerating integration of alternative models, accepting degraded functionality during a transition, or carving out narrower “low‑risk” workflows where Claude can continue operating under Anthropic’s guardrails.
Hegseth’s order, if formalized, would also cascade to cloud, chip, and defense‑IT partners that embed or train on Anthropic models. General‑purpose cloud providers that host Claude‑based services for federal clients are already reviewing whether they must ring‑fence those instances or unwind them entirely. The step has galvanized the sector: hundreds of employees at Google and OpenAI signed an open letter urging AI firms to align on red lines against domestic mass surveillance and fully autonomous weapons; retired cyber chiefs and lawmakers are urging narrower, rules‑based oversight to avoid chilling defense innovation while still preserving room for individual companies to refuse specific high‑risk uses.
The Pentagon’s rebrand-and its legal limits
Hegseth serves as “Secretary of War” under a September 5, 2025 executive order that authorized “Department of War” as a secondary title; under federal statute, the agency remains the Department of Defense unless Congress acts. The rebrand has symbolic and messaging effects, but procurement authorities and statutory naming remain anchored in existing law, including the broader U.S. defense acquisition system overseen by the Office of the Secretary of Defense and congressional defense committees. For contractors, the key point is that branding changes on social media or in speeches do not alter the underlying legal framework that governs contract awards, suspensions, and debarments.
- February 26, 2026 – Anthropic publicly rejects Pentagon demands to drop guardrails; Hegseth threatens Defense Production Act and “supply‑chain risk” actions.
- February 27, 2026 – Trump orders agencies to stop using Anthropic; Hegseth declares the firm a “supply‑chain risk” and directs contractors to sever ties within six months.
- March 2-3, 2026 – OpenAI amends its Department of War deal to bar domestic surveillance and autonomous weapons; former NSA chief Paul Nakasone criticizes the Anthropic designation as strategically short‑sighted.
As of March 3, 2026, Trump’s ban is in force with a six‑month off‑ramp for embedded uses. Hegseth’s supply‑chain‑risk action is being operationalized inside the acquisition system and faces expected legal challenge. Claude remains present in some defense systems pending migration, while OpenAI and other vendors move to fill classified demand under revised terms that are, for now, closer to the government’s preferred “any lawful purpose” standard than to Anthropic’s firmer red lines. The outcome of that legal and policy fight will help determine how much power frontier AI developers ultimately wield over the most sensitive decisions their tools enable.
Worth a look
