# Anthropic Government Ban On February 27, 2026, [[Donald Trump]] ordered every federal agency to immediately cease using [[Anthropic]]'s AI technology, and [[Pete Hegseth]] designated the company a "supply chain risk to national security" — an extraordinary use of procurement power to punish a private firm for refusing to drop [[AI Safety]] guardrails from its models. [[Anthropic]] sued, a federal judge called the ban "pretextual" [[First Amendment]] retaliation, and the case became the defining collision between the [[AI Arms Race]] for military dominance and the safety commitments that distinguish [[Anthropic]] from rivals like [[OpenAI]]. > [!important] Resolved on the merits — August 27, 2026 > The ban is **over**. Judge **Rita F. Lin** granted [[Anthropic]] **summary judgment** in a 59-page order, **vacated** both the supply-chain-risk designation and Hegseth's contractor-boycott order, entered a **permanent injunction**, and **denied** the government a stay. The sections below trace how the case got there; the outcome is in [The merits ruling](#the-merits-ruling-august-27-2026). ## Background: The Classified-Network Contract In July 2025 [[Anthropic]] and the [[Department of Defense]] signed a contract worth roughly $200 million that made Claude the first frontier AI model approved for use on classified networks — a landmark for both the company and the Pentagon's [[Defense Innovation Unit]]. Crucially, the [[Department of Defense]] agreed to abide by [[Anthropic]]'s acceptable use policy (AUP), which prohibited two categories of use: mass domestic surveillance of Americans, and fully autonomous weapon systems capable of selecting and engaging targets without human oversight. Those restrictions tracked [[Anthropic]]'s longstanding [[AI Safety]] commitments and sat comfortably alongside existing [[Department of Defense]] policy on [[Autonomous Weapons]], which already required meaningful human control over lethal force decisions. The contract was part of a broader defense-AI build-out that also included the [[Palantir-Anthropic-Amazon Defense AI Deal]], pairing [[Palantir]]'s data-integration platform with [[Anthropic]]'s models on [[Amazon]]'s cloud infrastructure. ## The Dispute Escalates By September 2025 the relationship had soured. Pentagon officials pushed for unrestricted "any lawful use" access to Claude — language that would have voided the AUP restrictions. [[Anthropic]] held firm: [[Dario Amodei]] publicly maintained that the company was willing to adapt its usage policies for the military but could not, given the current state of AI reliability, permit mass domestic surveillance or fully autonomous targeting. In January 2026 [[Pete Hegseth]] issued an AI strategy memorandum directing that all [[Department of Defense]] AI contracts incorporate standardized "any lawful use" language within 180 days. The memo was a direct shot at [[Anthropic]]'s contract terms and reflected Hegseth's broader push to strip safety restrictions from military AI procurement — a posture that mirrored the [[Trump Administration]]'s deregulatory stance across the tech sector. On the morning of February 27, Hegseth gave [[Dario Amodei]] a deadline of 5:01 PM that day to allow the [[Department of Defense]] unrestricted use of Claude "for all legal purposes." [[Anthropic]] refused. ## The Ban What followed was immediate and sweeping. [[Donald Trump]] directed all federal agencies to "immediately cease all use" of [[Anthropic]] products, with a six-month phase-out period for existing deployments. Hegseth simultaneously declared that no military contractor "may conduct any commercial activity with [[Anthropic]]" and formally designated the company a [[Supply Chain]] risk to national security — a classification ordinarily reserved for foreign adversaries and compromised vendors, not American AI companies headquartered in [[Silicon Valley]]. The General Services Administration removed [[Anthropic]] from USAi.gov, the federal government's centralized platform for agencies to test and procure AI models. Across departments from Defense to Health and Human Services, federal workers lost access to Claude overnight. ## OpenAI Steps In Hours after the ban, [[OpenAI]] announced it had signed a deal with the Pentagon to provide its AI tools on classified networks — timing that was widely read as opportunistic. [[Sam Altman]] posted on X that [[OpenAI]]'s contract included seemingly identical guardrails: "Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems. The DoD agrees with these principles, reflects them in law and policy, and we put them into our agreement." The symmetry undercut the [[Trump Administration]]'s rationale. If [[OpenAI]] could negotiate the same restrictions [[Anthropic]] had insisted on and still win a Pentagon contract, the ban looked less like national-security policy and more like selective punishment. The episode deepened questions about [[OpenAI]]'s relationship with the administration and the competitive dynamics of the [[AI Arms Race]] in the defense sector, where [[Palantir]], [[Anduril]], [[Scale AI]], and [[Shield AI]] were all positioning for military contracts under more accommodating terms. ## Anthropic's Legal Challenge On March 9, 2026, [[Anthropic]] filed two federal lawsuits — one in the U.S. District Court for the Northern District of California, one in the D.C. Circuit Court of Appeals — challenging the supply-chain risk designation and the government-wide ban. The complaints alleged [[First Amendment]] retaliation (the government punished [[Anthropic]] for its protected speech about [[AI Safety]]), exceeded the supply-chain statute's scope (the designation was designed for foreign threats, not domestic policy disagreements), and circumvented proper contract-cancellation procedures. CFO [[Krishna Rao]] submitted a court declaration laying out the financial damage: the government's actions could reduce [[Anthropic]]'s 2026 revenue by "multiple billions of dollars," with hundreds of millions at risk from [[Department of Defense]]-specific work alone. [[Amazon]], [[Google]], and [[Microsoft]] — [[Anthropic]]'s major cloud and investment partners — publicly reaffirmed their support for the company, signaling that the tech industry viewed the ban as a dangerous precedent. ## The Injunction On March 24, Judge Rita F. Lin of the Northern District of California held a hearing in which she stated bluntly that the ban "looks like punishment." Two days later she issued a sweeping preliminary injunction barring the [[Trump Administration]] from enforcing the presidential directive or the supply-chain risk designation. Her ruling found that "the record strongly suggests that the reasons given for designating [[Anthropic]] a supply chain risk were pretextual and that the real motive was unlawful retaliation" — language that situated the case squarely in [[First Amendment]] doctrine on government retaliation against private speech. Federal workers regained access to Claude. Agencies like the Department of Health and Human Services sent emails notifying staff that their Claude accounts, conversation histories, and data had been restored. ## Ongoing Legal and Regulatory Battle The victory was partial. On April 8, the D.C. Circuit Court of Appeals denied [[Anthropic]]'s motion for a stay in the parallel case, leaving the supply-chain risk designation formally in place for Pentagon contracting purposes even as the California injunction blocked its broader enforcement. [[Anthropic]] remained barred from new [[Department of Defense]] work. The [[Trump Administration]] also escalated on a second front: in June 2026, the Commerce Department classified [[Anthropic]]'s latest models as "cyber weapons" subject to export controls, making it illegal for [[Anthropic]] to provide them to any foreign national — including its own non-U.S. employees. The move was interpreted as further pressure on a company that had publicly defied the administration's defense-AI agenda. On July 31, 2026, a judge questioned the government's evidentiary basis for the supply-chain risk designation, finding that the administration had not provided sufficient proof that [[Anthropic]] poses an actual security threat. That hearing set up the summary-judgment posture resolved four weeks later. ## The merits ruling (August 27, 2026) **Judge Rita F. Lin granted [[Anthropic]] summary judgment** in a **59-page order**, on three independent grounds: | Claim | Holding | |---|---| | **[[First Amendment]]** | The measures were **retaliation** for [[Anthropic]]'s protected speech about [[AI Safety]] | | **[[Fifth Amendment]] due process** | Imposed **without the pre-deprivation process** the Constitution requires | | **[[Administrative Procedure Act]]** | **Outside the supply-chain statute's scope**, and **arbitrary and capricious** | The relief was total in substance: - **Vacated** the supply-chain-risk designation. - **Vacated Hegseth's boycott order** — the directive barring *military contractors* from doing any commercial business with [[Anthropic]]. This is the clause with the widest reach, and its vacatur is what unwinds the ban's effect on third parties. - **Permanent injunction** entered. - **Stay denied.** The government asked for a seven-day administrative stay; Lin refused, noting it had operated under her preliminary injunction since March without ever identifying a concrete harm. On the reasoning, Lin found the Pentagon's measures were rooted in a desire to **make a public example of [[Anthropic]] for its "arrogance"** in pushing back on the administration, rather than in any evidence the company intended to sabotage its own technology. Her line: "The empty invocation of national security is not a blank check to punish and retaliate against government critics." **[[Anthropic]] did not win everything.** Lin **rejected** the argument that [[Donald Trump]]'s directive exceeded presidential authority, and entered judgment **for** the agencies that had taken no action of their own. The holding is about *how* the executive acted — the process, the pretext, and the statutory scope — not about whether a president may direct agency AI procurement at all. ## What the ruling changes, and for whom The vacatur of the **boycott order** matters more widely than the designation itself, because it reached companies that were never party to the dispute. The most consequential third party is [[Palantir]]. The [[Maven Smart System]] runs [[Anthropic]]'s Claude inside it via [[Palantir AIP]], under the [[Palantir-Anthropic-Amazon Defense AI Deal]] — and Hegseth's order had barred defense contractors from doing business with [[Anthropic]] *at all*, whether or not the work was military. That put Palantir on a path toward an expensive model swap inside its flagship Pentagon program. The ruling removes that pressure. Palantir was not a litigant and takes the benefit incidentally. The shape of that is worth stating plainly: **a court found the government's national-security framing pretextual in a case brought by a company that refused a mass-surveillance carve-out, and the practical beneficiary is the surveillance-adjacent contractor whose product depends on that company's model.** ⚠️ **Open as of this writing:** the government's appeal posture is unreported. Note the split procedural history — the **D.C. Circuit denied a stay of the designation in April**, while the Northern District of California has now vacated it outright. The [[Anthropic Model Export Directive]] (the June 2026 Commerce "cyber weapons" export classification) is a **separate action on a separate track** and is not addressed by this ruling. ## Public Response and Market Effects The ban triggered an unexpected backlash. Claude surged to the top of the U.S. App Store as consumers downloaded the app in solidarity with [[Anthropic]]'s safety stance — the first time Claude had topped the charts. The episode became a rare instance of an [[AI Safety]] position generating mass public support rather than the usual niche policy-community interest. For the defense-tech ecosystem, the case sharpened a question the KB tracks across the [[Military-Industrial Complex]], [[Project Maven]], and the [[Defense Innovation Unit]]: can AI companies maintain meaningful safety commitments while competing for military contracts, or will the pressure to accommodate "any lawful use" demands erode those commitments industry-wide? [[OpenAI]]'s ability to negotiate identical restrictions while maintaining the administration's favor suggests the answer may depend less on the policy substance and more on the political relationship. ## Significance With the August 27 merits ruling, this stops being an open dispute and becomes **precedent**: the first judicial holding that procurement power used against an American company for its safety speech is First Amendment retaliation, that a supply-chain-risk designation cannot be stretched to cover a domestic policy disagreement, and that a contractor-wide boycott order imposed without process is void. Whether it survives appeal is the remaining question. The Anthropic Government Ban is the KB's live case study at the intersection of several major threads: the [[Trump Administration]]'s use of procurement power as political leverage, the tension between [[AI Safety]] and the [[AI Arms Race]], the [[First Amendment]] limits on government retaliation against corporate speech, and the competitive jockeying among [[OpenAI]], [[Anthropic]], [[Google]], [[Microsoft]], [[Palantir]], and [[Anduril]] for the defense-AI market. It is the first time a sitting administration has banned an American AI company from government use over a policy disagreement about safety guardrails — and the legal and political precedent it sets will shape the defense-tech landscape for years.