What happens when the question isn’t whether an AI tool works, but whether the government is allowed to say no to it?
That’s the question a U.S. appeals court in Washington just answered. In a 2-1 decision handed down Friday, the court upheld the Pentagon’s designation of Anthropic as a supply-chain risk. The practical effect: the Defense Department can keep barring Anthropic’s AI tools from its systems, and Anthropic stays shut out of Defense Department work.
If you follow AI mostly through product launches and benchmark scores, this one might feel like it came from a different universe. So let me unpack what’s actually going on here, and why it matters even if you never plan to sell software to the military.
What “supply chain risk” actually means
The phrase sounds like it belongs in a shipping manifest, and in a sense it does. A supply chain is just the network of companies and components that feed into a finished product or service. When the Pentagon buys software, that software becomes part of its supply chain. So do the vendors behind it, the infrastructure they run on, and the people who maintain it.
A “supply chain risk” designation is the government’s way of saying: we don’t want this in the pipeline. It’s not a criminal finding. It’s not a product review. It’s a procurement decision with teeth, and once it’s applied, the door closes.
The court’s reasoning centered on authority rather than on Anthropic’s technology. According to the decision, the law granting the Pentagon this power gave Defense Secretary Pete Hegseth wide latitude to make that call. In other words, the judges weren’t asked to grade Claude. They were asked whether the Secretary was allowed to make the judgment he made, and two of the three said yes.
Why the 2-1 split is the interesting part
A unanimous ruling reads like settled law. A 2-1 ruling reads like an argument that isn’t finished. One judge on that panel disagreed, which tells you there’s a real legal question here about how much room a Defense Secretary should have when deciding a private company poses a risk.
That tension is worth sitting with. Broad discretion makes sense when national security is on the line and speed matters. Broad discretion also means a single official’s judgment can effectively remove a major AI company from one of the largest buyers of technology on the planet, with limited recourse.
What this means if you’re not a defense contractor
Here’s the part I think matters for readers of this site. We usually talk about AI agents in terms of capability. Can it write the email? Can it read the spreadsheet? Can it book the meeting without hallucinating a date that doesn’t exist? Those are the questions that shape whether a tool is useful to you day to day.
This ruling is about a completely different axis: permission. Who is allowed to use which AI, and who decides?
For the last few years the AI conversation has been driven mostly by what the models can do. Decisions like this one point to a future where availability gets carved up by institution. Your employer’s rules, your industry’s regulator, your government’s procurement lists. An AI agent can be excellent and still be off-limits to you because of where you work.
A few things worth watching as this pattern develops:
- Institutional gatekeeping becomes normal. If the Pentagon can exclude a vendor with wide latitude, other large buyers, public and private, will notice the template.
- Vendor choice stops being purely technical. Organizations may pick AI providers based on which ones stay eligible for contracts, not which ones perform best.
- The rules are being written in courtrooms. Not in model cards, not in research papers. In appeals court decisions with split panels.
The uncomfortable simplicity of it
What strikes me most is how little this had to do with AI itself. The court addressed a question about statutory authority. The technology was almost incidental to the legal reasoning.
That’s a useful reminder for anyone trying to understand where AI is heading. The technical frontier gets the headlines, but the boundaries around these tools are being drawn by procurement law, agency discretion, and judicial review. Those systems predate the current wave of AI by decades, and they’re not going to reshape themselves around it.
Anthropic remains excluded from Defense Department work, and the Pentagon’s authority to make that call stands, at least for now. For the rest of us, the takeaway isn’t about one company’s contract prospects. It’s that the question “which AI should I use?” is quietly turning into “which AI am I permitted to use?” Those are not the same question, and only one of them is answered by how good the model is.
🕒 Published: