The Pentagon just made it official. Eight AI companies got the keys to America's classified networks on Friday. Anthropic was not one of them.

The Defense Department announced agreements with Amazon Web Services, Google, Microsoft, Nvidia, OpenAI, SpaceX, and Reflection AI to deploy their tech on classified military systems. Oracle was added a few hours later. The terms are blunt. Each company agreed to let the Pentagon use their AI for any lawful use, with no veto over operational decisions.

Anthropic refused that language. So the Pentagon refused Anthropic.

This is the resolution of a feud that has been running since February, when Defense Secretary Pete Hegseth declared Anthropic a supply-chain risk. We covered it then, we covered the Google classified deal that backfilled the gap, we covered Pentagon AI chief Michael confirming the freeze on camera, and we covered the White House quietly drafting a Mythos workaround so federal civilian agencies could still use Anthropic's flagship model. Friday was the moment all those threads collapsed into one sentence.

Anthropic is out. Everyone else is in.

The any lawful use line is the actual story. Reuters and the New York Times both confirmed it. The phrase comes straight from the contracts. It means the Pentagon, not the vendor, decides what the AI gets used for. Targeting decisions, surveillance, intelligence analysis, autonomous weapons coordination. All of it lawful, all of it the customer's call.

This is exactly what Anthropic has been refusing to sign for three months. Their public position is that AI labs need to draw red lines that vendors enforce. The Pentagon's position is that vendors do not write US military doctrine. One side had to blink. It was not the side with the budget.

CNBC interviewed Pentagon tech chief Michael on Friday. The exchange was not subtle. Anthropic, he said, is still blacklisted. Mythos is a separate issue. Translation: the White House guidance lets civilian agencies onboard Mythos through a regulatory tunnel, but the Defense Department contracts, the ones that move billions, stay off limits.

That is the worst possible split for Anthropic. The civilian carveout was supposed to be the win. It is now the consolation prize. The Pentagon door is welded shut for as long as the safety guardrail position holds, and it is going to hold because Dario Amodei has built the entire IPO pitch around it.

Now run the math on the IPO. Anthropic targets an October listing at $400 to $500 billion. The S-1 has to disclose material risk factors. The Defense Department blacklist is now permanent on the record. So is the missing federal classified line item. So is the fact that seven competitors, including the company Anthropic just sued in California for sabotaging the Mythos buildout, are now selling to the same customer Anthropic cannot reach.

Bankers do not love that page in the prospectus.

There is a counter-read, and Anthropic's PR team will push it hard this week. The any lawful use clause is a brand asset for safety-first enterprise customers. Banks, hospitals, law firms, anyone who needs the AI to refuse certain instructions will read these contracts and decide that Anthropic is the only major US lab still willing to refuse the customer. That is the exact pitch that Mythos sold into the federal civilian carveout. It will sell into the Fortune 500 too.

The Atlantic ran a piece this morning making the broader version of this argument. Anthropic grew four times as much last quarter as Google did during three years of its peak. The revenue line is not the problem. The problem is whether the safety-first brand still scales after the Pentagon publicly humiliates it.

Here is what to watch over the next 30 days. Whether any of the eight new Pentagon vendors poach Anthropic's federal civilian footprint. Whether the White House actually publishes the Mythos guidance text or lets it sit. Whether Anthropic's pre-IPO valuation in the secondary market holds at the trillion-dollar level Forge Global was printing two weeks ago, or starts retracing. And whether the trial across the bay in Oakland, where Sam Altman will eventually testify under oath about his own nonprofit pledge, gives Anthropic any cover to reposition.

The arms dealer always wins. Google sold $40 billion in capital to Anthropic and the Pentagon contract to itself. SpaceX got the IPO disclosure and the classified deal. Nvidia sold the chips. OpenAI got the federal lock-in two months early and is now selling Codex through AWS Bedrock at the same time.

Anthropic got a brand. We are about to find out what that is worth at $500 billion.