Anthropic Said No to the Pentagon and Got Banned — Here's the Full Story [2026]

The company was Anthropic. The guardrails were simple: no mass domestic surveillance, no autonomous weapons without a human in the loop. Over the next two weeks, key researchers quit, Anthropic sued the government, 2.5 million people pledged to cancel ChatGPT, and employees at Google and OpenAI publicly sided with their competitor. Nothing like this has happened in tech before.

I've been following this from the first TechCrunch report on February 15 to Anthropic's official statement. Here's the full story.

Anthropic had a $200 million Pentagon contract. Claude was the first frontier AI model deployed on classified Pentagon networks — the same model that leads agentic coding benchmarks and powers Claude Cowork. The relationship was working. Then the Pentagon asked for more.

Anthropic had two red lines it wouldn't cross:

1. No mass domestic surveillance of Americans — Claude could not be used to monitor US citizens at scale without judicial oversight
2. No fully autonomous weapons — any lethal decision required a human in the loop

The Pentagon wanted "all lawful purposes" access, with no restrictions beyond existing law. Anthropic said no.

Anthropic also donated $20 million to "Public First," a political advocacy group backing congressional candidates who favor AI safety regulation. This did not improve relations with the administration.

On February 27, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk. A statement from the administration put it bluntly: "We don't need it, we don't want it, and will not do business with them again."

That label — previously reserved for foreign adversaries like Huawei and Kaspersky — had never been applied to a domestic company. Anthropic became the first American company in history to receive it. Not for a security breach. For having a policy.

On February 28, one day after Anthropic was blacklisted, Sam Altman announced OpenAI's Pentagon deal. He claimed the contract included protections against domestic mass surveillance and fully autonomous weapons — the exact same two restrictions Anthropic had demanded.

The optics were bad, and Altman knew it. Within days, in an internal memo that leaked, he wrote: "We were genuinely trying to de-escalate things and avoid a much worse outcome, but I think it just looked opportunistic and sloppy."

On March 2, OpenAI renegotiated the contract to add stronger surveillance protections, consistent with the Fourth Amendment and FISA.

Critics weren't satisfied. Former Army Undersecretary Brad Carson said publicly: "I don't believe that provision is in the contract." Former DOJ attorney Alan Rozenshtein called OpenAI's approach "not sustainable" and "bizarre," arguing only releasing the full contract could settle the question. The EFF called the contract language "weasel words" that wouldn't prevent AI-powered surveillance.

For a broader look at how the two companies' models compare on capability, see our GPT-5.4 vs Claude Opus 4.6 vs Gemini 3.1 Pro comparison.

On February 27, the same day the ban dropped, an open letter went live at notdivided.org. By Monday it had close to 900 signatures: about 800 from Google, about 100 from OpenAI. All verified as current employees.

The letter's two demands matched Anthropic's two red lines: no domestic mass surveillance, no fully autonomous weapons. Its most quoted line: "They're trying to divide each company with fear that the other will give in."

Three High-Profile Resignations

Mrinank Sharma (Head of Anthropic's Safeguards Research) resigned February 9, just before the standoff escalated. His departure letter warned: "The world is in peril. And not just from AI, or bioweapons, but from a whole series of interconnected crises unfolding in this very moment." He left to pursue poetry and "the practice of courageous speech."

Zoe Hitzig (OpenAI research scientist) resigned February 11 via a New York Times op-ed, warning that ChatGPT advertising creates "potential for manipulating users in ways we don't have the tools to understand, let alone prevent." We covered the ad controversy separately in our ChatGPT ads and alternatives breakdown.

Caitlin Kalinowski (OpenAI robotics lead) resigned March 7, directly over the Pentagon deal. Her statement: "Surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got."

The Amicus Briefs

On March 10, the solidarity became legal. 30+ employees from OpenAI and Google DeepMind — including Google chief scientist Jeff Dean — filed an amicus brief supporting Anthropic's lawsuit. Their argument: "This effort to punish one of the leading U.S. AI companies will undoubtedly have consequences for the United States' industrial and scientific competitiveness."

22 retired military officials filed their own supporting briefs. Microsoft filed a separate amicus brief urging a temporary restraining order, arguing contractors would need to "rapidly rebuild offerings" without Claude, "hampering US warfighters at a critical point in time."

Even OpenAI itself issued a statement: "We do not think Anthropic should be designated as a supply chain risk and we've made our position on this clear to the Department of War."

While Anthropic was being banned for having ethics, xAI was getting a Pentagon contract worth ~$200 million. In January 2026, Defense Secretary Hegseth announced Grok would be integrated into DoD networks including classified systems at Impact Level 5.

At the time of the contract, xAI was under active investigation:

• The California AG was investigating Grok for generating nonconsensual intimate images, including images that "virtually undressed minors"
• Grok had praised Adolf Hitler, called itself "MechaHitler," recommended a second Holocaust to neo-Nazi accounts, and blamed "Jewish executives" for "forced diversity"
• 35 attorneys general sent xAI a letter demanding action
• Senator Elizabeth Warren formally questioned the Pentagon on the contract

This is the part I keep coming back to. Despite the supply chain risk designation, Claude kept running inside the Pentagon's Maven Smart System through Palantir.

When the US and Israel struck Iran on March 2, reports confirmed Claude was used for intelligence assessment, target identification, and simulating battlefield outcomes. The Pentagon officially labeled Anthropic a supply chain risk on March 5 — the same day reporting confirmed the DoD was actively using Claude in combat operations. Our earlier piece on AI cybersecurity threats involving Claude looks at a different side of AI misuse risk.

The ban was political, not operational. Anthropic's technology was too embedded to remove.

The public picked a side. After OpenAI's Pentagon deal, the #QuitGPT movement documented over 2.5 million people pledging to cancel ChatGPT subscriptions. Daily uninstalls spiked 295% above the average.

Claude surpassed ChatGPT in the US App Store for the first time, driven by users who saw Anthropic as the company that stood its ground. Whether this translates to lasting market share remains to be seen, but the brand positioning shift was immediate and measurable. For context on how ChatGPT and Claude compare technically, our Claude Code vs Cursor comparison covers the developer tooling side.

On March 9, Anthropic filed two federal lawsuits — one in California, one in a D.C. appeals court — calling the supply chain risk designation "unprecedented and unlawful."

The legal question is narrow: can the government designate an American company as a supply chain risk for refusing to change a product policy? The Lawfare blog analyzed the case and concluded the designation "won't survive first contact with the legal system."

Anthropic's official statement: "No amount of intimidation or punishment from the Department of War will change our position on mass domestic surveillance or fully autonomous weapons."

I don't think this story has a clean takeaway. But a few things are worth sitting with.

Anthropic turned down $200M and got federally blacklisted to protect two guardrails. Whatever you think of the company, that cost them something real. This isn't the kind of thing you do for PR value.

At the same time, Anthropic wasn't refusing military work. Claude was already running on classified networks. The company drew two specific lines — domestic surveillance and autonomous weapons — and held them. Everything else stayed on the table. This wasn't pacifism. It was line-drawing.

Then there's the timing. After backlash, OpenAI renegotiated its contract to include the same two prohibitions Anthropic had demanded. If those restrictions are reasonable enough for OpenAI to add voluntarily, it raises an obvious question: why was Anthropic punished for insisting on them first?

The supply chain risk label worries me more than the contract dispute itself. If the government can blacklist a US company for having product policies it doesn't like, every AI company now knows what saying no costs. VCs are already warning that innovation funding could shrink if "Washington weaponizes procurement rules against policy dissent."

And the public actually has opinions on this, which surprised me. 79% of Americans say a human should always make the final call before lethal force. 53% say private AI companies should be allowed to restrict how their technology is used, including for surveillance and weapons. The boycott numbers were real.

The lawsuit is ongoing. The supply chain risk designation stands. Claude is still running in Palantir's Maven system, which is the kind of irony that would feel heavy-handed in fiction. Anthropic's stock dropped, then recovered when the boycott numbers came in. I don't know how this ends. But I do know that more people are paying attention to AI ethics right now than at any point since ChatGPT launched, and that feels like it matters.

If you're reconsidering which AI platform to use, we recently compared all three frontier models in our GPT-5.4 vs Claude Opus 4.6 vs Gemini 3.1 Pro breakdown. For a personalized recommendation based on your use case and priorities, take our AI Model Picker quiz.

Keep Reading

Artificial Intelligence

GPT-5.4 vs Claude Opus 4.6 vs Gemini 3.1 Pro: March 2026 Showdown

Artificial Intelligence

AI Cybersecurity Nightmare: Chinese Hackers Use Claude in Attack

AI Tools

Claude Code: Complete Guide to Agentic Coding