OpenAI Signed the Pentagon Deal. Anthropic Wrote It.

Friday night, Sam Altman posted a statement so carefully calibrated it read like a legal brief disguised as a social media post. OpenAI had reached an agreement with the Department of War. Its models would enter classified networks. The deal included explicit protections against autonomous weapons and mass domestic surveillance.

Every one of those protections was something Anthropic had been demanding for months. And been punished for demanding.

Hours earlier, Defense Secretary Pete Hegseth had designated Anthropic a supply chain risk. The classification is historically reserved for foreign adversaries. Never before applied to an American company. President Trump ordered every federal agency to cease all use of Anthropic technology and threatened criminal charges if the company wasn't "helpful" during the six-month transition. Anthropic, the Pentagon's first AI partner in classified environments since June 2024, was now officially a national security threat.

Then OpenAI walked in and got the same deal. By its own telling, a better one.

The speed told the real story. The government blacklisted a leading AI company, installed a competitor within hours, and called both outcomes a win. The AI industry just learned how Washington plans to do business.

The terms Anthropic bled for

The standoff had been escalating for weeks. The Pentagon wanted unrestricted access to Claude across "all lawful uses." Anthropic held two red lines: no mass domestic surveillance of Americans, no fully autonomous weapons. Dario Amodei argued that current AI models aren't reliable enough for autonomous targeting, and that constitutional protections against surveillance shouldn't rest on a chatbot vendor's terms of service.

Hegseth set a 5:01 PM Friday deadline. Anthropic let it pass.

The government's response was punitive and personal. Hegseth called the company's position "a master class in arrogance and betrayal" and accused Anthropic of "defective altruism." Trump labeled them "Leftwing nut jobs" on Truth Social. Before midnight, with the wreckage still smoking, Altman announced his deal.

What OpenAI actually got

Read OpenAI's FAQ and you find a claim worth reading twice. The company says its contract provides "better guarantees and more responsible safeguards than earlier agreements, including Anthropic's original contract." OpenAI isn't saying it matched Anthropic's terms. It says it surpassed them.

The specifics are worth your attention. OpenAI's agreement is cloud-only. No edge deployment, no models running on drones or autonomous field systems. The AI stays in data centers. The company retains full control over its safety stack, including the ability to run and update classifiers independently of the Pentagon. Cleared OpenAI engineers will work inside the deployment. And the contract pins itself to current law, meaning that even if Congress later weakens surveillance protections or rewrites autonomous weapons policy, the Pentagon's use of OpenAI models still has to comply with today's legal standards.

That last provision is the one that matters. Anthropic never got anything close.

Fortune reported that the government agreed to let OpenAI refuse instructions its models won't follow. If accurate, then what exactly was the Pentagon fighting Anthropic over for three months? Anthropic asked for two narrow exceptions that, by its own account, hadn't affected a single government mission. OpenAI got a multi-layered enforcement architecture with forward-deployed engineers, cloud-only constraints, and a legal time lock. Same principles. Radically different packaging.

What Anthropic's refusal bought everyone else

Here is what nobody in this story wants to say out loud. OpenAI's deal exists because Anthropic refused to fold.

Look at the sequence. The Pentagon demanded unrestricted access. Anthropic said no. The government escalated publicly and theatrically, designating an American company a supply chain risk for the first time in the statute's history. Backlash came fast. About 500 engineers at OpenAI and Google put their names on an open letter. "We will not be divided," it read. Senators from both parties jumped in. Wicker, Reed, McConnell, Coons, all defense committee chairs, sent letters pushing both sides toward a deal. The Pentagon had spent weeks escalating. Now it looked cornered by its own noise.

OpenAI arrived with a proposal that gave the government everything it needed politically. A signed contract. A cooperative AI partner on classified networks. A visible win. The actual provisions, meanwhile, embedded protections that went further than anything Anthropic had managed to extract during months of hostile negotiation.

The Pentagon needed an exit ramp. Altman built one.

If you're an executive at any AI company watching this, the lesson stings. Anthropic's principled stand didn't save Anthropic. It changed the terms for everyone who came after. Holding the line cost Anthropic its Pentagon contract and its standing with the federal government. The better deal that emerged went to someone else. Anthropic wrote the deal. OpenAI signed it.

The real product OpenAI sold

The question everyone asked Friday night was blunt. If OpenAI's red lines are identical to Anthropic's, why did the Pentagon accept one company and reject the other?

Altman answered it himself, buried in a staff memo. "We believe this dispute isn't about how AI will be used, but about control. We believe that a private US company cannot be more powerful than the democratically-elected US government."

That framing is everything.

Anthropic drew a line and dared the government to cross it. That same line appeared in OpenAI's proposal, but packaged as a gift. Both companies arrived at similar contractual language. Only one made Hegseth feel like he'd won.

Forget capabilities. For classified deployment, Claude and OpenAI's models perform at roughly the same level. Posture closed this deal. Anthropic spent three months refusing to bend on language it considered existential, anxious about every precedent a concession might set. Altman waited, then offered the Pentagon identical terms wrapped in cooperative language and a public willingness to "de-escalate."

Undersecretary Jeremy Lewin went on X with the official line. OpenAI got "the same compromise that Anthropic was offered, and rejected." But OpenAI's published contract includes cloud-only deployment, a retained safety stack, and time-locked legal baselines that Anthropic's earlier agreement apparently lacked.

Same deal, or a better one? Both. That's the tell.

Who collects from here

OpenAI walks away with the Pentagon contract and a $110 billion funding round announced the same day, valuing the company at $840 billion. Altman positioned himself as the industry's diplomat, the executive who could work with Washington. Amazon Web Services announced the same week it would host OpenAI's models, removing the last technical barrier to classified cloud deployment.

Anthropic walks away blacklisted, its $380 billion valuation under pressure, Polymarket projections sliding. But the company built something that doesn't fit on a term sheet. Nearly 500 engineers at rival companies publicly backed Anthropic's position. Nate Silver, not exactly an activist, wrote that OpenAI "increasingly looks like the Facebook/Meta of the AI world, cutthroat and very bottom-line focused." When your competitor's own employees sign open letters defending you, that's a kind of capital no funding round can buy.

Silver raised another question that should worry Altman's recruiters. Claude has caught up to ChatGPT despite starting with considerably fewer resources. That gap looks like culture and talent, exactly the kind of advantage that erodes when your own engineers start backing your competitor's ethics.

And the precedent? Grim. The government showed it will punish companies for negotiating in the open and reward those who arrive later with cooperative optics. If you're Google, Meta, or any lab watching this week, the playbook writes itself. Let someone else absorb the hit. Walk in after.

The next company that says no

Anthropic will challenge the supply chain risk designation in court. The legal argument looks strong. The statute targets foreign adversaries, and applying it to a domestic company over contract terms would make every government vendor a potential hostage. Whether courts move fast enough to change anything on the ground is a separate question.

But the more revealing test comes next time. The next capability the Pentagon wants broad access to. The next company deciding whether to hold the line publicly or work the angles privately.

OpenAI's deal proves the guardrails were always available. The Pentagon was never going to put ChatGPT on a drone. The fight was always about who gets to say so, and how loudly. Anthropic said it loud and got labeled a national security threat. OpenAI barely had to say anything. It just collected the check.

Hegseth needed to break one company before he could sign the next. He got both done before midnight.

Anthropic Says No. Google Gives It Away. Congress Says Nothing.

San Francisco | Friday, February 27, 2026 The Pentagon sent its best and final offer on Claude military use. Dario Amodei's answer: no. Anthropic will lose $200 million in defense contracts and acces

The Implicator

OpenAI's Pentagon Deal Claims the Same Red Lines That Got Anthropic Blacklisted

Sam Altman announced Friday night that OpenAI reached an agreement with the Pentagon to deploy its AI models on classified military networks, claiming the deal preserves the same safety red lines that

The Implicator

OPINION: Anthropic Drew the Line. Congress Should Have Drawn It First.

On Thursday, Dario Amodei rejected the Pentagon's "best and final offer" to lift restrictions on military use of his company's AI model Claude. "We cannot in good conscience accede to their request,"

The Implicator