What if the biggest factor deciding which AI assistant you end up using at work has nothing to do with how good the AI is?
That is the uncomfortable idea sitting underneath a court ruling that landed this week. A federal judge found that the Trump administration’s blacklisting of Anthropic, the company behind the Claude AI models, was illegal. The Pentagon’s move to sideline the company was ruled unlawful. Several outlets covered it, including The Guardian, Politico, The New York Times, The Wall Street Journal, and Ars Technica, with Ars noting the blacklisting was tied to the company being labeled “woke.”
I write about AI agents for people who do not build them, and my instinct with a story like this is to ask a simple question: does this change anything for me, the person who just wants software that works? I think it does, and not in the way most coverage suggests.
Why a procurement fight matters to normal people
Most of us pick AI tools the way we pick a coffee shop. We try one, it works, we keep going back. We assume the market sorted out the options before they reached us, and that the good ones rose to the top.
Government procurement does not work like that. When a large buyer, and the Pentagon is one of the largest buyers of anything on earth, decides a vendor is off the table, that decision does not stay inside the building. It moves through the supply chain. Contractors adjust. Partners hedge. Integrators who build agent systems for government clients quietly stop recommending the blacklisted option, because why introduce a risk into a proposal?
The AI agent you use two years from now may be shaped less by which model reasons best and more by which vendors survived the political weather. That is a strange thing to accept, and I do not think we should accept it quietly.
The part that should make you pause
Set aside how you feel about Anthropic specifically. Set aside how you feel about the administration. Focus on the mechanism.
A company was excluded from a buyer’s list. A court found that exclusion illegal. That sequence tells us something about how fragile vendor selection can be when the reasons for it are not written down in a way that survives review. Reporting points to the company’s perceived politics as the trigger, which is a rationale that has nothing to do with whether the software is accurate, safe, or well built.
For anyone choosing AI tools inside an organization, this is a useful reminder that “approved vendor” and “good vendor” are two different lists. They overlap most of the time. They are not the same list.
What this means if you are picking AI tools
You are probably not procuring for the Pentagon. But you might be the person at your company who decides which AI assistant the team uses, or which agent gets access to your customer data. A few practical takeaways:
- Do not treat a vendor list as a quality ranking. Inclusion and exclusion can happen for reasons that have nothing to do with the product. Ask what criteria were used.
- Build for portability. If your workflows are wired so tightly to one model provider that switching would take months, you have taken on a risk you did not choose. Keep your prompts, data, and logic separable from any single model.
- Watch who is making the decision. When politics enters procurement, the loudest signal is often not about the technology at all. Learn to tell those signals apart.
- Expect the rules to move. A court can reverse a ban. A new administration can create one. Plan as if your vendor list will change, because it will.
A less comfortable read
There is an optimistic version of this story where the courts act as a check, the illegal ban gets reversed, and the system works as designed. That reading is fair. A judge looked at a government decision and said no.
The less comfortable reading is that the ban existed at all, and that a company needed a court to undo it. Legal correction takes time, and in that time, markets move. Contracts get signed. Teams commit to architectures. A ruling can restore a company’s eligibility without restoring the opportunities that passed by while the matter was in dispute.
For a technology that its own builders describe as one of the most consequential of the century, the idea that access to it can turn on a vendor’s perceived politics is worth sitting with.
The takeaway for the rest of us
AI agents are becoming infrastructure. They read your email, draft your documents, and increasingly take actions on your behalf. Infrastructure choices are sticky, and they tend to outlive the reasoning that produced them.
So when you next hear that a particular AI tool is or is not allowed somewhere, ask why. The answer may be technical. It may be about safety or accuracy or cost. Or it may be about something else entirely, and this ruling is a reminder that “something else entirely” is a real category.
Knowing the difference is not a technical skill. It is a literacy one, and it is becoming part of the job.
🕒 Published: