Trump's AI Thaw: From Pentagon Blacklist to Possible Anthropic Deal

President Donald Trump surprised observers Tuesday. A deal letting Anthropic's AI models run on Pentagon networks? "It's possible," he told CNBC's Squawk Box. Just weeks after blacklisting the company as a national security risk. Anthropic CEO Dario Amodei had met White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent on April 18. Trump called the talks "very good." "They're shaping up," he said. "Very smart. They can be of great use."

The turnaround caps a bruising fight. Back in July 2025, Anthropic inked a $200 million contract. Its Claude models would deploy on classified Defense Department networks -- the first frontier AI to do so. Negotiations soured by September. The Pentagon demanded unfettered access for "all lawful purposes." Anthropic held firm on two red lines: no mass domestic surveillance, no fully autonomous weapons.

Deadline hit February 27, 2026, at 5:01 p.m. No deal. Trump fired off a Truth Social post: Federal agencies must "IMMEDIATELY CEASE" using Anthropic tech. Defense Secretary Pete Hegseth followed minutes later on X. Anthropic? A "supply chain risk to national security." No military contractor could touch it. "America's warfighters will never be held hostage by the ideological whims of Big Tech," Hegseth wrote, per CNN.

OpenAI pounced. Hours later, CEO Sam Altman announced a Pentagon pact. Models for classified nets, with safeguards. Rival filled the void fast, as NPR reported. Anthropic? Revenue hit $30 billion annually anyway. But the blacklist stung. Military ties severed. IPO dreams clouded.

Courts jumped in. U.S. District Judge Rita Lin ruled March 2026. The ban smacked of "classic illegal First Amendment retaliation." Preliminary injunction blocked enforcement across agencies, according to The Next Web. D.C. federal appeals court pushed back April 8. Upheld the risk label. Legal limbo.

Mythos changes the equation.

Anthropic's new model. Cybersecurity powerhouse. Spots vulnerabilities. CISA tests it. Intelligence community too. An administration source told Axios: "It would be grossly irresponsible for the US government to deprive itself of the technological leaps that the new model presents. It would be a gift to China." Trump echoed that. Meeting productive, per White House. Anthropic confirmed "shared priorities."

But hurdles loom. Pentagon stance unchanged. No formal risk-label withdrawal. Office of Management and Budget drafts access protocols. Lobbying helps: Anthropic hired Ballard Partners. Wiles's old firm. Political savvy.

And the stakes? Huge. AI arms race. China looms. U.S. needs edge. Anthropic's Claude powered Iran strikes via Palantir's Maven, Washington Post revealed. Blacklist locked out that capability. Mythos? Even sharper. Trump wants it back. "We want the smartest people," he said on CNBC, as CNBC covered.

Industry watches. OpenAI entrenched. But Anthropic's safety stance won allies. Catholic theologians filed briefs. Moral tradition meets Big Tech. Revenue surges despite ban. Now, possible truce.

Will it stick? Trump feigned ignorance at first. "Who?" on Amodei. "I had no idea." Classic. Yet talks advanced. Deal teases strategic pivot. Pentagon gets Mythos muscle. Anthropic regains access. China stays behind.

Or not. Courts grind on. Hegseth silent. One thing clear. AI controls national security. Companies draw lines. Government pushes back. Hard.

Trump's words signal thaw. Possible. Not done. Watch this space.