Anthropic built Claude to be helpful, harmless, and honest — and it placed two limits on that helpfulness: Claude would not be used for mass surveillance of Americans, and it would not be deployed in fully autonomous weapons systems. Those two limits are now the centre of a public standoff between the company and the U.S. Department of Defense.
Claude holds a distinction no other AI model has: it is the first large language model to be deployed on the U.S. military’s classified network. That deployment came with a contract worth $200 million. On Tuesday, February 24, 2026, Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to the Pentagon and issued a direct ultimatum — remove the restrictions and allow Claude to be used “for all lawful purposes,” or lose the contract. The Pentagon also threatened to designate Anthropic a “supply chain risk,” a label the U.S. government typically reserves for companies connected to foreign adversaries.
Amodei refused. By Thursday, February 26, he published a detailed statement holding Anthropic’s position. The Pentagon had set a hard deadline of 5:01 p.m. ET on Friday, February 27. Anthropic did not move.
We cannot in good conscience accede to their request.
— Dario Amodei, CEO, Anthropic (February 26–27, 2026)Unfolding in the same week — and reported separately by TIME on February 23, 2026 — was a major internal change at Anthropic. The company overhauled its Responsible Scaling Policy (RSP), scrapping the original 2023 commitment to halt AI training if safety measures were not already in place. The two developments — a public stand on usage limits and a quiet revision of internal safety commitments — ran just days apart, offering a fuller picture of where Anthropic stands heading into 2026. For context on the broader AI chip and compute race shaping these decisions, see our coverage of the Nvidia earnings, tariffs, and Claude Code security story.
A chronological look at every key event — hover or tap each card for detail.
Tap a perspective to see what was at risk — and why each position was held.
Anthropic’s Position & What Was at Risk
- What it refused: Removing two specific restrictions — no mass surveillance of Americans, no fully autonomous weapons use.
- Direct financial risk: A $200 million contract — significant in absolute terms, but modest against a $380 billion valuation and annualised revenue growing at 10× per year.
- Bigger exposure — supply chain label: Being classified a “supply chain risk” could disrupt commercial partnerships far beyond the single DoD contract.
- Talent and trust: Amodei argued capitulating would erode trust from safety-focused researchers who chose Anthropic specifically for its stated values.
- No operational disruption claimed: Amodei stated the two restrictions had not slowed “adoption and use of our models within our armed forces to date.”
- For context on Claude’s commercial rise, see our Claude Code and AI chip coverage.
The Pentagon’s Demands & Leverage
- Core demand: Use Claude “for all lawful purposes” — without a private company imposing restrictions on military decision-making.
- Three tools of pressure: Contract cancellation; supply chain risk designation; and potential invocation of the Defense Production Act — a Cold War-era law that could compel access to the technology.
- Pentagon’s stated position: DoD spokesman Sean Parnell wrote the military has “no interest in using AI to conduct mass surveillance of Americans (which is illegal)” and does not want “autonomous weapons that operate without human involvement.” Critics noted this directly contradicted the demand for unrestricted use.
- Parallel strategy: The Pentagon was simultaneously negotiating with OpenAI and Google for the same terms. An open letter from AI workers warned companies were being played against each other.
- Contradictory threats: Amodei pointed out that designating Anthropic a security risk while simultaneously invoking emergency powers to gain access to Claude’s capabilities are logically incompatible positions.
What the AI Industry Watched Closely
- OpenAI, Google, and Elon Musk’s xAI all hold separate military AI contracts and were in parallel negotiations with the Pentagon for the same concessions.
- Workers from OpenAI and Google signed a public open letter backing Anthropic — an unusual cross-company statement on an active commercial dispute.
- The open letter stated: “The Pentagon is negotiating with Google and OpenAI to try to get them to agree to what Anthropic has refused. They’re trying to divide each company with fear that the other will give in.”
- The dispute re-opened the 2018 Project Maven debate, where Google ultimately declined to renew its contract after employee protests and pledged not to use AI in weaponry.
- Explore the broader AI infrastructure investment race for the commercial backdrop to these negotiations.
Wider Concerns Raised
- The two contested use cases are distinct: mass surveillance raises civil liberties concerns; fully autonomous weapons raise questions about human oversight of lethal force.
- Ret. Lt. Gen. Jack Shanahan — inaugural director of the DoD’s Joint Artificial Intelligence Center (JAIC) — stated LLMs are “not ready for prime time in national security settings,” specifically for autonomous weapons. He called Anthropic’s red lines “reasonable.”
- Chris Painter of METR (an AI model evaluation nonprofit) warned the RSP v3.0 shift shows Anthropic “believes it needs to shift into triage mode with its safety plans, because methods to assess and mitigate risk are not keeping up with the pace of capabilities.”
- Painter added: “This is more evidence that society is not prepared for the potential catastrophic risks posed by AI.”
- The Trump administration has resisted federal AI regulation — no binding national AI law is in place. Open-source AI ecosystem initiatives represent one alternative governance direction being explored.
Direct quotes from official statements, blog posts, and verified social media posts only.
Those latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security.
Dario Amodei CEO, Anthropic — public statement, Feb 26, 2026 AnthropicIt’s a shame that @DarioAmodei is a liar and has a God-complex. He wants nothing more than to try to personally control the US Military and is ok putting our nation’s safety at risk.
Emil Michael Undersecretary for Research & Engineering, DoD — post on X PentagonSince I was square in the middle of Project Maven & Google, it’s reasonable to assume I would take the Pentagon’s side here. Yet I’m sympathetic to Anthropic’s position. More so than I was to Google’s in 2018.
Ret. Lt. Gen. Jack Shanahan Inaugural Director, DoD Joint AI Center (JAIC) — social media post IndependentPainting a bullseye on Anthropic garners spicy headlines, but everyone loses in the end.
Ret. Lt. Gen. Jack Shanahan Inaugural Director, DoD JAIC — social media post, Feb 26, 2026 IndependentThis is more evidence that society is not prepared for the potential catastrophic risks posed by AI.
Chris Painter Director of Policy, METR — statement to TIME IndependentTime and time again over my three year tenure at Anthropic I’ve seen us stand to our values in ways that are often invisible from the outside. This is a clear instance where it is visible.
Trenton Bricken Technical Alignment Team, Anthropic — post on X AnthropicSeparate from the Pentagon dispute, Anthropic’s chief science officer Jared Kaplan confirmed to TIME that the company overhauled its Responsible Scaling Policy. RSP v3.0 was published on February 24, 2026. Here is what changed:
- Would not train a new AI model unless safety measures were proven adequate in advance — a hard pre-condition on commercial release
- Binary “tripwire” system: capability thresholds would automatically trigger a development pause
- Categorical bar on training above a defined power level without pre-existing mitigations
- Positioned as a potential blueprint for national regulation and international treaties
- Removes unilateral training-halt commitment — will not pause if competitors advance without equivalent safeguards
- Regular public Frontier Safety Roadmaps with concrete safety goals
- Risk Reports every 3–6 months: capabilities, threat models, and mitigations — with independent expert review for the most advanced models
- Will delay development only if Anthropic leads the AI race and sees significant catastrophic risk — both conditions must apply
- Commits to matching or surpassing competitors’ safety efforts
If one AI developer paused development to implement safety measures while others moved forward training and deploying AI systems without strong mitigations, that could result in a world that is less safe. The developers with the weakest protections would set the pace, and responsible developers would lose their ability to do safety research.
— Anthropic RSP v3.0 introduction, approved unanimously by Dario Amodei and Anthropic’s board, published February 24, 2026. Source: anthropic.comPost-money valuations at confirmed, publicly announced funding rounds only.
Sources: Anthropic Series F · Anthropic Series G · Sacra research. Only officially announced valuations shown.
Tap any question to expand. Each answer is based on verified statements and reporting only.
Anthropic’s two restrictions — no mass surveillance, no fully autonomous weapons — remained in place as the Pentagon’s 5:01 p.m. ET Friday deadline passed. No agreement was announced. Amodei’s statement described the company’s willingness to support a transition to another provider if the DoD proceeded, and noted that the two restrictions had not caused operational disruption to existing military use of Claude.
RSP v3.0, published on February 24, 2026, was approved unanimously by Amodei and Anthropic’s board. Chief Science Officer Jared Kaplan described the change as a pragmatic response to a regulatory environment that did not materialise and a competitive landscape that did not pause. The revised policy was reviewed in draft form by METR. The new version retains commitments to transparency through public Frontier Safety Roadmaps and quarterly Risk Reports, including independent external expert review for the most advanced models.
The Pentagon standoff was reported from February 24–27, 2026. The TIME RSP exclusive was published February 23, 2026 — four days before the deadline. Both developments were discussed across the industry, in open letters from AI workers at competing companies, and in public statements from Republican and Democratic lawmakers. Related developments in the AI sector are tracked at Giganectar’s AI coverage.
