Top Pentagon Official Rejects Anthropic Researcher’s AI Warning

A top Pentagon official is pushing back against rising fears about artificial intelligence risks days after a public warning from an Anthropic PBC engineer that AI insiders believe the technology “could kill us all by the end of the decade.”

Emil Michael, the Defense Department’s chief technology officer and top research and engineering official, referred to ex-Anthropic engineer Jacob Coxon’s social media post — currently at more than 150 million views — in remarks on Thursday that touched on fears of a “dystopian” world dominated by unstoppable AI systems.

“It’s easy to get caught up in this sort of doom loop of, ‘We’re going to unemploy 40% of Americans, we’re going to destroy humanity,'” Michael said at a defense industry conference in Washington. “It just goes to fear of data centers, fear of change — all the fears that have coalesced in one well-written tweet.”

In a series of posts earlier this week, Coxon warned that Open AI and Anthropic were “racing straight to self-improving superintelligence and gambling with our lives,” and that there “will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.”

Michael, a former senior executive at Uber Technologies Inc., has spearheaded Pentagon efforts to accelerate AI adoption across the US military. He said the potential harms from AI, like those highlighted by Coxon’s post, could be mitigated by the free market, as well as through industry-wide collaboration and engagement with the government.

The Defense Department’s AI push touched off a bitter dispute with Anthropic over the firm’s demands for extra safeguards on military use of its AI tools. Michael helped lead negotiations with the company, but those talks broke down earlier this year after the Pentagon rejected Anthropic’s terms, calling them an infringement on the military’s authority, and declared the firm a threat to the U.S. supply chain.

The conflict has shifted to federal court, where Anthropic has sued in both California and in Washington, D.C., seeking to overturn the Pentagon’s designation. The judge in the California lawsuit has ruled in the company’s favor, though the federal appeals court in Washington has yet to issue a decision. Michael called the D.C. Circuit case “more important” and insisted that Anthropic remains a supply-chain risk in the absence of a preliminary injunction.

Anthropic did not immediately respond to a request for comment about Michael’s remarks on Thursday.

He accused the company of having “forum-shopped” for litigation in the politically liberal Northern District of California, calling the outstanding case in the DC Circuit “more important.” Pointing to the likelihood of a decision in the coming months from what he termed “more of a national security-oriented forum.”

The department has now transitioned about 90% of its classified AI workload away from Anthropic and is on track to shift the rest by the end of this month, Michael later told reporters. “All of the Maven Smart Systems or Palantir work has been transitioned months ago,” he said, referring to the prominent defense tech firm’s AI system.

Despite Commerce Secretary Howard Lutnick saying Anthropic was now back on side following a “kerfuffle” with the administration, Michael drew a distinction between the Pentagon’s national security needs and those of the rest of the federal government.

“I’m glad that they fixed their issues with the rest of the government,” Michael said, noting that the Defense Department “has been very consistent about our need to be working with AI providers in a way that benefits the war-fighter, the defense of the nation. That means we need to use software if we’re going to use it — for all lawful purposes, in an unconstrained way.”