Updated February 28, 2026: The situation resolved — and revealed a striking contradiction. On February 27, Anthropic was formally blacklisted. Hours later, OpenAI announced a Pentagon classified network deal containing the same two redlines (no mass surveillance, human control over autonomous weapons) Anthropic was blacklisted for insisting upon. Full reporting below.

Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei until 5:01pm on Friday, February 27, 2026 to agree to terms for Pentagon use of Claude models — or face Defense Production Act compulsion and designation as a “supply chain risk.” Anthropic’s response, published the day before the deadline, was an unambiguous public refusal.

The February 24 Meeting

According to the Financial Times, Hegseth summoned Amodei to Washington on February 24, 2026. A senior Pentagon official stated:

“Anthropic had until 5.01pm on Friday ’to get on board or not’.”

The specific measures threatened, per the same official:

MeasureDescription
Supply chain exclusionCut Anthropic from Pentagon procurement
Defense Production ActCompel usage “regardless of if they want to or not”
Risk designationLabel Anthropic “a supply chain risk”

“If they don’t get on board, [Hegseth] will ensure the Defense Production Act is invoked on Anthropic, compelling them to be used by the Pentagon regardless of if they want to or not.” — Senior Pentagon official (via FT)

Anthropic’s Position

Anthropic has maintained usage policy restrictions on certain military applications. Per the FT, citing people familiar with the negotiations, these include:

  • Lethal missions without human control
  • Mass domestic surveillance
  • Unfettered classified access

Anthropic’s statement to FT:

“Good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s national security mission in line with what our models can reliably and responsibly do.”

Anthropic’s Public Response: The “Department of War” Statement (February 26, 2026)

One day before the compliance deadline, Dario Amodei published a formal public statement. The title was pointed: “Statement from Dario Amodei on our discussions with the Department of War.”

The “Department of War” was the official name of what is now the Department of Defense — until 1949. The naming choice was not accidental.

Opening:

“I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries.”

The Two Redlines

Anthropic confirmed two specific conditions it would not remove:

RedlineStated Reason
Mass domestic surveillance“AI-driven mass surveillance presents serious, novel risks to our fundamental liberties… incompatible with democratic values.” Anthropic distinguishes this from foreign intelligence and counterintelligence — which it supports.
Fully autonomous weapons“frontier AI systems are simply not reliable enough to power fully autonomous weapons… We will not knowingly provide a product that puts America’s warfighters and civilians at risk.” Partially autonomous systems — cited in the context of Ukraine — remain permissible.

What Anthropic Does Support

The statement clarifies Anthropic is cooperating with military and intelligence use across several areas:

  • Foreign intelligence and counterintelligence operations
  • Partially autonomous weapons systems
  • Classified network access — Anthropic claims to be the first frontier AI provider deployed in U.S. classified networks and National Laboratories

Holding the Line

On the threats themselves:

“Regardless, these threats do not change our position: we cannot in good conscience accede to their request.”

Anthropic stated willingness to “facilitate transition to alternative providers if offboarded” while keeping current model access available under its proposed terms.

The Logical Contradiction

Amodei noted the inherent contradiction in the DOD’s dual threats: one labels Anthropic a security risk requiring exclusion; the other treats Claude as so essential it justifies legal compulsion under the Defense Production Act. Both cannot be simultaneously true.

Revenue Sacrifice as Precedent

The statement cited Anthropic foregoing “several hundred million dollars in revenue” to restrict Claude access to CCP-linked firms — framing consistent values-driven policy decisions as a track record, not a one-off.

Source: Dario Amodei, “Statement from Dario Amodei on our discussions with the Department of War” — February 26, 2026

Existing Pentagon Relationships

Anthropic currently has defense sector relationships at risk:

RelationshipValueContext
DOD Contract$200 millionExisting agreement
Palantir PartnershipClassified missionsClaude has been sole cleared model
Maduro OperationJanuary 2026Claude deployed in capture operation

The January 2026 Maduro operation prompted queries from Anthropic about usage specifics, per FT sources.

Defense Production Act Context

The Defense Production Act (1950) grants presidential authority to direct domestic industry for national defense. Historical applications:

  • COVID-19: Medical supply production (Trump/Biden administrations)
  • Critical minerals: Domestic production increase (Trump administration)
  • Anthropic AI: Potentially first AI provider compulsion

If invoked, Anthropic would be legally required to provide access without consent.

Other Provider Positions

ProviderStatusSource
xAI/Grok“On board with being used in a classified setting”Pentagon official via FT
OpenAI✅ Deal signed — classified network deployment, redlines acceptedAltman X post, Feb 27, 2026
GoogleNegotiatingFT

Political Context

The deadline occurs within a broader policy divergence:

  • Anthropic position: Advocates for AI safety regulation, warns of existential risk (policy history)
  • Administration position: Trump administration promotes “light-touch regulatory framework”

Trump AI tsar David Sacks (who divested xAI positions upon appointment) called Anthropic “woke” in October 2025, accusing the company of “running a sophisticated regulatory capture strategy based on fear-mongering.”

Why the Redlines Matter: Real-World Stakes

The debate over “autonomous weapons” and “mass surveillance” is not abstract. Human rights groups have documented AI targeting systems already deployed in active conflicts. Israel’s use of AI in Gaza has been widely reported, naming systems described by sources as “Lavender,” “The Gospel,” and “Where’s Daddy?” — tools reportedly used for surveillance, target generation, and tracking individuals. These are the practical applications that gave rise to the specific language in both Anthropic’s and OpenAI’s redlines.

The concern isn’t hypothetical. When AI providers negotiate whether to remove prohibitions on “autonomous lethal weapons” or “mass surveillance,” these are the use cases under discussion. The DOD’s demand that Anthropic accept “any lawful use” — while simultaneously negotiating with OpenAI to formally embed the same prohibitions — suggests the dispute was not purely about safety policy.

Timeline

DateEvent
January 2026Claude used in Maduro capture operation
February 24, 2026Hegseth-Amodei meeting in Washington
February 25 AMAnthropic announces RSP v3 update
February 25 PMFT publishes deadline story
February 26, 2026Amodei publishes “Department of War” statement; holds both redlines publicly
February 27, 2026 5:01pmCompliance deadline
February 27, 2026Trump orders all federal government to cease using Anthropic; Hegseth formally designates Anthropic a supply chain risk
February 27, 2026300+ Google / 60+ OpenAI employees sign “We Will Not Be Divided” solidarity letter
February 27, 2026Anthropic vows legal challenge; second statement calls designation “legally unsound”
February 27, 2026Sam Altman announces OpenAI-DOD classified network deal — with the same two safety redlines

Note on the OpenAI timing: The OpenAI deal was announced on the same day as the blacklisting. Altman’s X post and the “Scaling AI for Everyone” blog were published February 27, 2026 — hours after the Anthropic designation was confirmed. Whether the timing was coordinated or coincidental is not confirmed.

February 27: The Deadline Outcome

Anthropic did not comply. The threats were carried out.

Defense Secretary Hegseth confirmed the supply chain risk designation on February 27, with Trump issuing a directive for the entire federal government to cease all use of Anthropic’s products. Hegseth’s statement:

“In conjunction with the President’s directive for the Federal Government to cease all use of Anthropic’s technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security.”

Key terms of the designation:

TermDetail
Contractor prohibitionNo contractor, supplier, or partner doing business with the U.S. military may conduct commercial activity with Anthropic
Transition periodAnthropic continues providing services for up to six months “to allow for a seamless transition to a better and more patriotic service”
Historical precedentBelieved to be the first time the designation has been applied to an American company — it is typically reserved for foreign adversaries

Anthropic responded the same day with a second statement, calling the designation “legally unsound” and pledging to challenge it in court:

Anthropic vowed to “challenge any ‘supply chain risk’ designation in court,” arguing it would set “a dangerous precedent for any American company that negotiates with the government.”

The company also disputed Hegseth’s interpretation: the supply chain risk designation legally applies to contractor relationships with the Pentagon, not to all commercial activity broadly — narrowing its actual enforcement scope.

Source: Anthropic: Statement on Comments by the Secretary of War — February 27, 2026

The Contrast: OpenAI Struck a Deal Hours Later — With the Same Redlines

Within hours of Anthropic’s blacklisting, OpenAI CEO Sam Altman announced a deal with the Department of Defense to deploy AI models in the DoD’s classified network. The announcement was part of OpenAI’s “Scaling AI for Everyone” post on February 27, 2026 — the same announcement that disclosed the largest venture capital deal ever recorded.

The funding: OpenAI closed a $110 billion round on February 27, 2026, setting the company’s post-money valuation at $840 billion — the largest venture deal ever recorded. Investors included SoftBank ($30B), NVIDIA ($30B), and Amazon ($50B). The Pentagon deal and the funding announcement dropped simultaneously.

The safety terms Altman described are nearly identical to the conditions Anthropic was blacklisted for insisting on. Altman’s X post:

“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 DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement.”

Compare to Anthropic’s two redlines:

RedlineAnthropic’s PositionOpenAI’s PositionDoD Response
Mass domestic surveillanceRefused to remove restrictionSame prohibition statedAccepted (OpenAI deal)
Autonomous weaponsRefused to remove human-control requirementSame prohibition on “human responsibility for use of force”Accepted (OpenAI deal)

The DOD accepted from OpenAI the exact safety limits it refused to accept from Anthropic — and blacklisted Anthropic for insisting upon.

The structural difference, if any, may be in language: OpenAI’s language (“human responsibility for the use of force”) versus Anthropic’s (“fully autonomous weapons, lethal missions without human control”). The DOD’s demand of Anthropic was to allow “any lawful use” with all safeguards removed. Whether the DOD accepted genuinely equivalent terms from OpenAI, or weaker language that preserved the political appearance of safety limits, is not yet publicly verifiable from the agreement text.

Industry Solidarity: “We Will Not Be Divided”

On February 27, 2026, more than 300 Google employees and over 60 OpenAI employees signed an open letter backing Anthropic’s position, titled “We Will Not Be Divided.” The letter called on company leadership to:

“Put aside their differences and stand together to continue to refuse the Department of War’s current demands for permission to use our models for domestic mass surveillance and autonomously killing people without human oversight.”

Company responses were cautious but sympathetic:

  • Sam Altman told OpenAI staff he wanted the company “to help de-escalate things,” adding: “We have long believed that AI should not be used for mass surveillance or autonomous lethal weapons.”
  • Jeff Dean (Google DeepMind Chief Scientist) publicly expressed opposition to mass domestic surveillance.

Neither Altman nor Dean signed the employee letter — but Altman’s announced deal reflected the same substantive limits the letter demanded.

Policy Context:

Risk Assessment:

Provider Alternatives:

Verification Ledger

✅ VERIFIED: High Evidence

Deadline and threats

  • Source: Financial Times, February 25, 2026
  • Quote: “Anthropic had until 5.01pm on Friday ’to get on board or not’”
  • Quote: “ensure the Defense Production Act is invoked on Anthropic, compelling them to be used by the Pentagon regardless of if they want to or not”
  • Quote: “label Anthropic ‘a supply chain risk’”

Financial details

  • Source: FT
  • Quote: “$200mn contract with the department”
  • Quote: “summoned… to Washington for a meeting on Tuesday”

Provider positions

  • Quote: “Musk’s Grok ‘is on board with being used in a classified setting’”
  • Quote: “Claude tool has… been the only model working on classified missions as a result of its partnership with Palantir”
  • Quote: “Claude was used in the US capture of Venezuelan leader Nicolás Maduro in January”

Anthropic’s public response (primary source — direct company statement)

  • Source: Dario Amodei, Anthropic CEO — February 26, 2026
  • URL: anthropic.com/news/statement-department-of-war
  • Quote: “I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries.”
  • Quote: “AI-driven mass surveillance presents serious, novel risks to our fundamental liberties… incompatible with democratic values.”
  • Quote: “frontier AI systems are simply not reliable enough to power fully autonomous weapons… We will not knowingly provide a product that puts America’s warfighters and civilians at risk.”
  • Quote: “Regardless, these threats do not change our position: we cannot in good conscience accede to their request.”
  • Verified: Two explicit redlines (mass surveillance, fully autonomous weapons) publicly confirmed
  • Verified: Anthropic supports foreign intelligence, counterintelligence, partially autonomous weapons
  • Verified: Anthropic claims to be first frontier AI in U.S. classified networks and National Laboratories
  • Verified: Revenue sacrifice cited — “several hundred million dollars” foregone to restrict CCP-linked access
  • Date verified: 2026-02-26

“Department of War” naming

  • Source: Statement title — Dario Amodei, February 26, 2026
  • Result: Deliberate editorial choice; “Department of War” was the DoD’s name until 1949
  • Scope: Framing is Anthropic’s own; no independent confirmation of intent, but the label has no other plausible reading

⚠️ PARTIAL: Attribution-Based

Anthropic’s specific concerns (pre-statement reporting)

  • Source: FT citing “people familiar with the negotiations”
  • Scope: Superseded in part by Anthropic’s direct Feb 26 statement
  • Update: Lethal missions, mass surveillance, unfettered classified access were confirmed in substance by the public statement

Legal action possibility

  • Source: FT citing “people familiar with the matter”
  • Scope: Not confirmed by Anthropic — statement does not mention litigation

February 27 designation — confirmed

  • Source: Hegseth statement + multiple outlets (The Hill, TechCrunch, CBS News, Washington Times, NPR)
  • Quote (Hegseth): “In conjunction with the President’s directive for the Federal Government to cease all use of Anthropic’s technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security.”
  • Note: DPA was not separately invoked — supply chain risk designation was the mechanism used
  • Date verified: 2026-02-27

Contractor prohibition and transition period

  • Source: Multiple outlets citing Hegseth designation text
  • Result: No military contractor may conduct commercial activity with Anthropic; 6-month transition period granted
  • Historical note: First time designation applied to an American company

Anthropic legal challenge

  • Source: Anthropic second statement (anthropic.com/news/statement-comments-secretary-war)
  • Quote: Vowed to “challenge any ‘supply chain risk’ designation in court,” calling it “legally unsound”
  • Legal argument: Designation legally applies only to Pentagon contractor relationships, not all commercial activity
  • Date verified: 2026-02-27

OpenAI DOD deal — same redlines accepted

  • Source: Sam Altman X post, Feb 27, 2026; OpenAI “Scaling AI for Everyone” blog post
  • Quote: “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 DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement.”
  • Significance: DOD accepted from OpenAI the same safety limits they blacklisted Anthropic for insisting upon
  • Date verified: 2026-02-27

OpenAI $110B funding round — post-money valuation $840B

  • Source: OpenAI “Scaling AI for Everyone” blog post, Feb 27, 2026; Al Jazeera, Feb 28, 2026
  • Amount: $110 billion — largest venture deal ever recorded
  • Investors: SoftBank ($30B), NVIDIA ($30B), Amazon ($50B)
  • Post-money valuation: $840 billion (pre-money $730B + $110B investment)
  • Timing: Announced simultaneously with the Pentagon classified network deal
  • Date verified: 2026-02-27/28

Employee solidarity letter

  • Source: TechCrunch, Engadget, CNBC, Yahoo News
  • Result: 300+ Google employees + 60+ OpenAI employees signed “We Will Not Be Divided” letter
  • Altman response: “We have long believed that AI should not be used for mass surveillance or autonomous lethal weapons”
  • Date verified: 2026-02-27

Sources

Primary Sources

Secondary Sources (Deadline Outcome)

Published: 2026-02-25 Last updated: 2026-02-28 Evidence level: High (primary Anthropic + OpenAI statements + multi-outlet confirmation)