The Pentagon awarded classified-use AI contracts to seven vendors on May 1, 2026 — OpenAI, Google, Microsoft, Amazon, Nvidia's peer xAI, Nvidia itself, and the startup Reflection — while pointedly leaving out Anthropic, the company that previously held a $200 million deal to handle classified materials for the agency. The Defense Department labeled Anthropic a supply-chain risk, the formal designation that triggered a federal ban on its products earlier this year. Anthropic sued in response and won a temporary injunction, but Friday's announcement makes clear the agency is routing classified workloads around it.
The agreements cover what the Pentagon calls the 'lawful operational use' of frontier AI systems across classified environments, framed as a step toward 'establishing the United States military as an AI-first fighting force.' Microsoft and Amazon already have deep cloud relationships with the department. The new entrants are Nvidia and Reflection. OpenAI and xAI had previously reached separate agreements for 'lawful' use of their systems, and The Information reported Google struck a similar deal in the run-up to Friday.
Anthropic's exclusion is the culmination of a months-long fight. The company refused to loosen its usage policies around mass domestic surveillance and fully autonomous weapons for Pentagon use, the red lines its safety team has held since founding. That refusal cost it the $200M classified contract and triggered the supply-chain-risk designation that locked its models out of federal procurement.
Key facts
- 01The Pentagon signed classified-use AI deals with OpenAI, Google, Microsoft, Amazon, Nvidia, xAI, and Reflection on May 1, 2026.
- 02Anthropic was excluded after the Defense Department labeled it a supply-chain risk.
- 03Anthropic previously held a $200M Pentagon contract for handling classified materials.
- 04Anthropic sued the federal government over the ban and won a temporary injunction.
- 05DoD CTO Emil Michael called Anthropic's Mythos model a 'separate national security moment' for cyber defense.
What makes the situation stranger is that the Pentagon is not done with Anthropic's technology entirely. Defense Department CTO Emil Michael told CNBC on Friday that Anthropic 'is still a supply chain risk' but called its security model Mythos a 'separate national security moment,' adding that 'we have to make sure that our networks are hardened up, because that model has capabilities that are particular to finding cyber vulnerabilities and patching them.' The NSA reportedly already has access to Mythos despite the broader ban.
“Anthropic had a $200 million deal to handle classified materials for the Pentagon before refusing to loosen its red lines on mass domestic surveillance and fully autonomous weapons.”— Jaeden Schafer
The political temperature around Anthropic specifically has run hot. Defense Secretary Pete Hegseth on April 30 called Anthropic CEO Dario Amodei an 'ideological lunatic,' a striking attack on the head of a company the federal government had been paying nine figures weeks earlier. President Trump struck a softer tone on April 21, saying 'it's possible' Anthropic and the Pentagon could reach a deal eventually.
For OpenAI, Google, and Nvidia, the upside is straightforward. Classified-use authorization is the gating credential for the largest single AI buyer in the world, and the deal pulls them onto the same procurement footing as Microsoft and Amazon, which have spent years building cleared infrastructure. Reflection's inclusion alongside that group is the surprise — a startup placed in a tier most defense contractors wait a decade to enter.
xAI's inclusion lands at an awkward moment for Elon Musk, who is currently being sued by OpenAI's co-founders in a trial in Oakland and conceded on the stand last month that xAI used OpenAI models to train Grok. None of that has slowed the Pentagon's appetite for his model. The agency wants optionality across vendors, and xAI offers a politically aligned alternative to the West Coast incumbents.
The competitive picture has shifted fast. Anthropic, which two weeks ago gave investors 48 hours to commit to a $50B round at a $900B valuation, is now formally locked out of the largest defense AI procurement of the year. Its commercial trajectory is intact — the company has been signing enterprise and creative partners, including the Claude Design launch into Canva and Figma — but the federal revenue line just got cut.
The counter-case for Anthropic is that holding the line on autonomous weapons and mass surveillance is exactly the brand it has spent four years building, and its investors priced the $900B round knowing the safety posture was non-negotiable. Mythos still has a federal foothold via the NSA. And the temporary injunction it won leaves a legal door open. Trump's April 21 comment that a deal is 'possible' suggests the freeze is not necessarily permanent.
What Friday really illustrates is that the AI safety debate has stopped being theoretical and started costing real contracts. Anthropic drew a line on autonomous lethality and domestic surveillance, the Pentagon drew a line on vendors who won't move theirs, and the gap between the two is now worth at least $200M and a seat at the table on every classified workload the U.S. military runs for the next several years. The other six vendors made a different choice, and the market just rewarded it.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
