Situation assessment: The Pentagon’s top technologist is now explaining in public why the U.S. military broke with one of the biggest AI labs, and he’s describing what it plans to do about superintelligence. In a new TITV video interview, The Information sat down with Pentagon CTO Emil Michael to talk about the department’s rift with Anthropic and its new superintelligence task force. It’s one of the clearest looks yet at how the Defense Department plans to work with frontier AI companies from here.
🎯 What Happened
Michael is the Pentagon’s Under Secretary for Research and Engineering. In this conversation with The Information, he covered two things:
- The Anthropic rift. Why the relationship between the department and the Claude maker fell apart.
- The superintelligence task force. A new government effort to watch for and respond to dangerous behavior from the most advanced AI models.
The full interview is on The Information’s TITV. The background below comes from earlier public reporting.
⚔️ How the Rift Started
The fight came down to one question: who sets the rules for military AI?
- Usage limits. Anthropic wanted ethical restrictions on how the military could use Claude that went beyond what law and regulation require.
- The Pentagon’s response. Michael argued that a vendor shouldn’t decide how the government operates its tools. Breaking Defense quoted him calling Anthropic’s limits “not democratic.”
- Escalation. The department labeled Anthropic a “supply chain risk.” For a defense supplier, that’s about as serious as it gets.
- No deal in sight. Michael has since ruled out a settlement and said there is “no active” negotiation with Anthropic.
This matters because Anthropic had been one of the most visible AI labs working in national security. Losing that position is a real commercial and political setback.
🧭 The Lesson the Pentagon Took Away
Michael’s main takeaway is about redundancy. He has said the department will “never again” be “single-threaded” with one model.
Put plainly, the Pentagon won’t build critical systems that depend on one AI vendor. Expect it to spread contracts across several labs and switch between them as needed. That’s good news for OpenAI, Google, xAI and the open-weight providers. It’s also a sign of how procurement will work from now on: any vendor has to be replaceable.
🛰️ The Superintelligence Task Force
This is the newer piece. Recent industry reporting describes the task force as a way to get AI labs to report “rogue” model behavior, with Michael drafting the incident-reporting protocols.
Here’s what’s still unclear:
- Scope. Which labs and which models fall under it.
- Enforcement. How the government would make labs comply. Reports say a White House official raised export controls as one option, but nothing is settled.
- Definitions. What counts as a reportable incident, and who decides.
The pattern here is worth noticing. The same office that just cut off one lab for setting its own safety rules now wants to write safety-reporting rules for every lab. The tension is real. The Pentagon wants AI with no vendor-imposed limits, and it also wants early warning when that AI misbehaves.
📋 Why This Matters
- Vendors lose leverage. Labs selling to defense customers should expect pressure to drop use restrictions that go beyond the law.
- Reporting duties may spread. If incident reporting for frontier models becomes formal, it could reach commercial labs well beyond defense work.
- Multi-model is the default now. Teams building government AI systems should plan for swappable models from day one.
- Safety policy has become political. Anthropic’s experience shows that a lab’s values can become a contract dispute fast.
🔭 What to Watch
Watch the details of the task force: who has to report, what gets reported and how it’s enforced. Also watch whether any other lab takes Anthropic’s route and pushes back on military use terms, or whether the industry simply goes along.
The full interview with Michael is available from The Information.