Emil Michael, the Under Secretary of Defense for Research and Engineering, said it plainly at a Washington conference this week: "We were single-threaded on one vendor, one AI vendor at the Department of War. But never again will we be single-threaded with any one model." [1]

As the paper's May 12 account described the civilian door opening while the Pentagon door stayed shut, the frame was Anthropic navigating a gap between institutions — a gap that Michael's "never again" statement has now named and hardened into doctrine.

He was talking about Anthropic. He was also describing a new procurement doctrine.

At roughly the same time, the White House was circulating internal drafts of policy plans that would permit civilian federal agencies to use Anthropic's Mythos cybersecurity model — a different answer to the same company from a different part of the same government. [2]

As the paper's May 12 account described the civilian door opening while the Pentagon door stayed shut, the frame was Anthropic navigating a gap between institutions. What Wednesday made clear is that the gap is now a named institutional split — not a transition or a negotiation, but two different settled positions held simultaneously by the Pentagon and the White House.

Michael's Statement and What It Means

At the Special Competitive Studies Project's AI+ Expo in Washington, Michael announced recent agreements with eight major technology companies: Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, Reflection, Oracle, and SpaceX. He characterized these deals as a "counterstatement" — a demonstration that the Pentagon can build AI capability without Anthropic. [3]

The "never again single-threaded" formulation is not a tactical grievance. It is a declared procurement philosophy. Michael is saying that the Pentagon's prior dependence on Anthropic — its Claude models underpinned significant DoD AI infrastructure before the supply chain risk designation — was a structural mistake, and one he intends to prevent from recurring with any vendor.

The context of the dispute is by now well-documented. The Pentagon designated Anthropic a supply chain risk — a designation previously reserved for foreign adversaries — after the company refused to permit its AI models for autonomous weapons and certain surveillance operations. Anthropic's acceptable-use policy drew a line the Pentagon characterized as operationally incompatible. Michael's view, stated earlier this year, was that the DoD needed an "all lawful use" standard that Anthropic would not accept.

When asked directly whether he saw Anthropic's situation with the government being resolved, Michael answered: "Not at the Department of War, no." [4]

The White House's Different Answer

Trump's public posture toward Anthropic is warmer than Michael's by a wide margin. "They're very smart," Trump said recently, "and I think they can be of great use. I think we'll get along with them just fine." [5]

The civilian-agency guidance in draft form would allow departments outside the Pentagon — agencies without the DoD's autonomous weapons equities — to access Anthropic's Mythos model, a cybersecurity tool capable of detecting decades-old vulnerabilities in browsers, infrastructure, and software. Mythos is the product that reopened White House engagement with Anthropic after the Pentagon dispute; it sits outside the categories that triggered the original conflict.

The result is two doors in the same building. The Pentagon door is closed and Michael says it will stay closed. The White House door is open, or opening — depending on whether the draft civilian-agency guidance becomes final policy.

Anthropic's Position

Anthropic has not publicly responded to Michael's "never again" statement. The company is in an unusual position: simultaneously blacklisted by one major government client, re-engaged by another part of the same government, and operating under the constraint of a legal challenge to the original supply chain risk designation — a federal judge halted the government's orders to remove Anthropic products from agencies pending litigation in late March. [6]

The civilian-agency path, if it becomes policy, gives Anthropic a foothold in federal government that does not require resolving the Pentagon dispute. Whether that foothold is enough — commercially, reputationally, as a precedent for future government relationships — depends on how large the civilian-agency AI market is relative to defense contracts.

The answer, measured in dollar terms, is that defense contracts are larger. But a company can be commercially viable without the Pentagon's business. The question is whether it can be commercially viable while being publicly designated a supply chain risk by the Pentagon, regardless of what the White House thinks.

Michael's "never again" is designed, in part, to make that question harder to answer with yes.

-- SAMUEL CRANE, Washington