Skip to content

Trump Bans Anthropic from Federal Use — The Dispute and the Reality of OpenAI's Deal

Audience: Business professionals and engineers following AI industry developments

Key Points

  • Federal government bans Anthropic Pentagon designates Anthropic a "supply chain risk" and orders all agencies to stop use
  • Dispute centers on surveillance and weapons clauses Anthropic demanded contractual use restrictions, arguing current law is insufficient for the AI era
  • OpenAI's deal terms remain opaque Altman claims similar principles, but the full contract is not public and cannot be independently verified

Important note

Defense-related agencies have been given a 6-month phase-out period. This is not an immediate, across-the-board shutdown.

On Friday, February 27, 2026, President Trump ordered all federal agencies to stop using Anthropic's technology1. Defense Secretary Pete Hegseth designated the company a "national security supply chain risk" and barred all military contractors from doing business with Anthropic2. (Note: This article uses "Pentagon" to refer to the Department of Defense, which the current administration has rebranded as the "Department of War.")

That same evening, OpenAI CEO Sam Altman announced a deal to deploy models on the Pentagon's classified networks3. On the surface, this looks like "Anthropic out, OpenAI in." But a closer look at the reporting reveals a more complicated picture.

The core question: Did OpenAI accept the same terms Anthropic refused? This article pieces together the facts from multiple primary sources.

Timeline of Events

DateEvent
July 2025Anthropic signs a contract worth up to $200M with the Pentagon. Claude becomes the first AI model cleared for use on military classified networks4
~Feb 24 (Tue)Defense Secretary Hegseth and Anthropic CEO Amodei meet at the Pentagon5. No agreement reached
Feb 26 (Thu)Pentagon issues ultimatum: accept terms by Friday 5:01 PM ET or the contract is terminated6
Same eveningAmodei issues a public statement: "We cannot in good conscience accede"7
Feb 27 (Fri) afternoonTrump posts on Truth Social ordering all federal agencies to stop using Anthropic (with a 6-month phase-out for defense-related agencies)1
Same day, 5:01 PM ETPentagon deadline passes. No agreement
Post-deadlineHegseth designates Anthropic a "supply chain risk"2. GSA removes Anthropic from USAi.gov and the MAS procurement framework8
Same eveningAltman announces OpenAI's deal with the Pentagon3

With this timeline established, let's examine the heart of the dispute.

What Was the Dispute About?

The conflict centers on two use restrictions Anthropic sought to include in the contract:

  • A ban on use for mass domestic surveillance of US citizens
  • A ban on use in fully autonomous weapons (weapons that execute attacks without human judgment)

The Pentagon wanted the right to use the technology for "all lawful purposes." Pentagon officials argued that "current law and DoD policy already prohibit surveillance and autonomous weapons. There's no need to put it in the contract"9.

Anthropic countered that "current law hasn't caught up with the AI era." Since AI can analyze publicly available data — social media posts, location data — at massive scale, substantive mass surveillance is possible even within the bounds of existing law10.

So how did OpenAI handle the same sticking points?

Did OpenAI Refuse or Accept the Same Terms?

This is the most commonly misunderstood point — and the most important one.

Altman's Claims

In his announcement, Altman stated the following3:

  • A ban on mass domestic surveillance and human responsibility for the use of force are OpenAI's most important safety principles
  • The Pentagon agreed to these principles and embedded them in law and policy, incorporated into the agreement
  • Technical safeguards (such as cloud-only model deployment) will also be built11
  • Researchers with security clearances will be stationed at the Pentagon to monitor usage11

Altman also said he "asked the Pentagon to offer these same terms to every AI company"3.

How Does This Differ from Anthropic's Terms?

According to Axios, the restrictions in OpenAI's deal reflect existing US law and Pentagon policy and were not intended to create new legal standards11. Anthropic, on the other hand, was arguing that existing law is inadequate and was seeking contractual constraints that go beyond what the law requires10.

There is one particularly notable piece of evidence. An Anthropic spokesperson said of the Pentagon's final contract language: "There was little progress on mass surveillance and autonomous weapons prevention. The language framed as compromise was paired with legalese that would allow those safeguards to be disregarded at will"12.

The picture that emerges from reporting looks like this:

Anthropic's demandsOpenAI's deal (per Altman)
Surveillance & weapons banStrict contractual constraints (current law insufficient)Claims principles are reflected in "law and policy" and the agreement
Accepting "all lawful purposes"Refused (law hasn't caught up to AI)Unknown (full contract not public)
Pentagon discretionConstrainedUnknown (Pentagon says it will operate within "law and policy," but enforceability of contract terms is not public)

Note

Neither company's full contract is public, and no third-party verification of the differences is possible. The comparison above is based on reporting from Axios, Reuters, CNN, and others — it represents the "reporting-based understanding," not established fact. Reuters noted the differences from Anthropic's red lines were "not immediately clear"13, and CNBC similarly sought clarification from both companies and the Pentagon14.

It is possible that OpenAI accepted contract terms Anthropic deemed unenforceable. But whether OpenAI secured substantively equivalent protections — or merely different wording with a different reality — cannot be determined at this time.

Industry Reaction

This situation extended well beyond Anthropic alone, rippling across the entire AI industry. Three developments stand out.

1. Open Letter: "We Will Not Be Divided" (Google & OpenAI Employees)

A joint letter signed by employees of both Google and OpenAI urged leadership to "set aside differences and stand united in supporting Anthropic's position"15. Signature counts varied by outlet and collection time:

  • Thursday 5:30 PM PT: 160+ Google, 40+ OpenAI15
  • Friday: 300+ Google, 60+ OpenAI16
  • Friday afternoon: 550+ total17

2. Internal Letter to Jeff Dean (Google)

Separately, over 100 members of Google's AI team submitted an internal letter to Chief Scientist Jeff Dean (per NYT reporting)18, demanding red lines on military use of Gemini. Similar internal pushback against unrestricted military AI use was reported at Microsoft and Amazon19.

3. Open Letter to the Pentagon and Congress (Against Retaliation)

A separate open letter signed by AI founders, investors, and engineers argued that "the government should not retaliate against a private company for declining to accept changes to a contract." Signatories included OpenAI researchers Boaz Barak and William Feng, and Waymark CEO Alexander Persky-Stern20.

Industry solidarity is clear, but how it affects Anthropic's business is a different question.

Business Impact on Anthropic

  • Valued at $380 billion with an annualized run rate of $14 billion4
  • The Pentagon contract (up to $200M) is limited relative to overall revenue
  • However, the "supply chain risk" designation could cascade across all military-adjacent business relationships, making the full impact unclear
  • An IPO has been widely expected, though the company has not made a final decision4. The effect on investor sentiment and other deals is hard to gauge
  • Anthropic has questioned the legal basis of the designation and signaled it may pursue legal action21

The Fundamental Question

This episode raises a question that will define the next era of AI:

Does a company that develops AI have the right to place conditions on how its technology is used?

In the traditional defense industry, companies almost never restricted use. But AI is not a conventional weapon — depending on how it is deployed, it can affect the very foundations of democracy. The two issues Anthropic raised — mass surveillance and autonomous weapons — sit squarely on that boundary.

The administration says "existing law is sufficient." Anthropic says "the law hasn't caught up to AI." Which side is right will ultimately be decided by future legislation and case law. But one thing is already clear: this is a debate every AI company will soon be forced to confront. The joint employee letters from Google and OpenAI are an early signal.

This article is based on reporting available as of February 28, 2026. Contract details remain non-public, and assessments may change as more information becomes available.

Post this article Copy link

  1. Trump, Truth Social post, Feb. 27, 2026. Full text available via CBS News and NPR. ↩↩

  2. Pete Hegseth, post on X, Feb. 27, 2026. Cited by NPR and CNN. ↩↩

  3. Sam Altman, post on X, Feb. 27, 2026. Full text cited by CNBC. ↩↩↩↩

  4. NPR, "President Trump bans Anthropic from use in government systems," Feb. 27, 2026. Contract value, valuation, run-rate revenue, and classified network clearance. NPR's original wording: "the first to be cleared for classified use." ↩↩↩

  5. CNN, "Trump administration orders military contractors and federal agencies to cease business with Anthropic," Feb. 27, 2026. On the Tuesday Hegseth-Amodei meeting. ↩

  6. Sean Parnell (Pentagon spokesperson), post on X, Feb. 26, 2026. The "5:01 PM ET" deadline. Cited by CBS News and CNN. ↩

  7. Dario Amodei, statement, Feb. 26, 2026. "Cannot in good conscience accede." Cited by PBS NewsHour (via AP). ↩

  8. GSA removal from USAi.gov and MAS (Multiple Award Schedule). Reported by CNN. Primary source: GSA official press release. ↩

  9. Emil Michael (Pentagon Under Secretary for Research and Engineering), interview with CBS News, Feb. 26, 2026. Statement that "current law and DoD policy already prohibit surveillance and autonomous weapons." Cited by CBS News and OPB/NPR. ↩

  10. Amodei's statement and Anthropic's position. TechCrunch, "Anthropic vs. the Pentagon: What's actually at stake?" Feb. 27, 2026. Details how AI analysis of publicly available data can constitute de facto surveillance within existing legal bounds. ↩↩

  11. Axios, "Pentagon approves OpenAI safety red lines after dumping Anthropic," Feb. 27, 2026. On "within existing law," "no new legal standards," cloud-only deployment, and security-cleared researchers. ↩↩↩

  12. Anthropic spokesperson, statement. "New language framed as compromise was paired with legalese that would allow those safeguards to be disregarded at will." Cited by CBS News and The Hill. ↩

  13. Reuters reporting. Stated differences from Anthropic's red lines were "not immediately clear." ↩

  14. CNBC. Noted the differences between OpenAI's contract and Anthropic's demands were unclear, and sought clarification from both companies and the Pentagon. ↩

  15. Axios, "Open letter urges Google and OpenAI to join Anthropic's red lines," Feb. 27, 2026. 160+ Google, 40+ OpenAI as of Thursday 5:30 PM PT. ↩↩

  16. TechCrunch, "Employees at Google and OpenAI support Anthropic's Pentagon stand in open letter," Feb. 27, 2026. 300+ Google, 60+ OpenAI. ↩

  17. Washington Post, Feb. 27, 2026. 550+ as of Friday afternoon. Engadget reported 450+ (~400 Google) at a different collection time. ↩

  18. The New York Times reporting. Referenced by Fortune: "more than 100 workers at Google sent a letter to Jeff Dean." ↩

  19. Fortune, "Trump orders U.S. government to stop using Anthropic," Feb. 27, 2026. On internal movements at Microsoft and Amazon. ↩

  20. NBC News, "OpenAI announces Pentagon deal after Trump bans Anthropic," Feb. 27, 2026. On the open letter to Congress and the Pentagon. "We strongly believe the federal government should not retaliate against a private company for declining to accept changes to a contract." ↩

  21. Anthropic, statement, Feb. 27, 2026. Called the designation "legally unsound" and "a dangerous precedent." Cited by CBS News and CNBC. CNN (via AFP) used the phrase "vowed to sue over the 'intimidation.'" ↩