What if the government could blacklist a company for refusing to help it watch people?
That’s not a dystopian hypothetical anymore. It nearly happened to Anthropic, the AI safety-focused startup behind Claude. And last month, a federal judge said: absolutely not.
On March 16, the Northern District of California issued a preliminary injunction that stopped the Department of Defense from designating Anthropic as a “supply-chain risk”—a bureaucratic label that functions like a corporate death sentence. No government contracts. No partnerships. No oxygen. The kicker? The Pentagon slapped that designation specifically because Anthropic insisted on contractual language that would prohibit its AI from being used for mass surveillance.
Let that sink in. The government tried to economically strangle a company for not building oppressive technology.
Why This Feels Like a Watershed Moment
Here’s what makes this case so electric: it’s the first real legal test of whether companies have the right to refuse to weaponize their own technology—and whether governments can retaliate when they do.
Anthroplic’s position was refreshingly clear. The company wasn’t being difficult or woke-washing. It literally wrote into its contracts that its models shouldn’t be used for “large-scale surveillance that harms civil liberties.” That’s it. A straightforward ethical guardrail.
The Pentagon’s response? Blacklist them.
The court ruled that Anthropic likely had a valid First Amendment claim—suggesting that refusing to participate in surveillance isn’t just a preference, it’s potentially protected speech.
What happened next is what makes this genuinely interesting. The court didn’t just slap the DOD’s hand. The judge signaled something larger: that Anthropic probably has a winning First Amendment argument here. Think about what that means. Under this logic, a company’s decision to not build something, or to not participate in something, can be constitutionally protected expression.
Is This Actually a Victory for Responsible AI, or Just Legal Theater?
Look, I’m bullish on this ruling, but let’s not pretend it’s the end of the story. A preliminary injunction isn’t a final judgment. It just means the court said “hold on, the Pentagon probably overstepped here, so don’t do this while we figure it out.” The full case hasn’t been decided yet.
But here’s what’s genuinely novel: for years, the AI industry has gotten away with a kind of moral sleepwalking. Build the capability, and let the government (or whoever) decide how to use it. That’s been the implicit contract. Anthropic broke it—deliberately—and the law just said that act has teeth.
This also creates interesting pressure on every other AI lab. OpenAI, Google, Meta—they’ve all been trying to position themselves as responsible actors while simultaneously pitching their models to every government agency with a budget. Now there’s legal precedent suggesting you might not be able to have it both ways.
What Happens to the Pentagon’s Case Now?
The preliminary injunction is temporary. It buys Anthropic time while the underlying legal question gets decided. The DOD will almost certainly appeal, and they might try to reclassify the designation under different legal language.
The real tension here is constitutional, not corporate. Can the federal government use its contracting power as a cudgel to punish speech it dislikes—or in this case, speech refusals? The First Amendment has something to say about that, and the judge seemed persuaded.
But here’s the unsettling part: the Pentagon has infinite resources and infinite patience. They can fight this for years. Anthropic can survive a preliminary injunction; it can’t survive a decade-long legal war. So the real question isn’t whether Anthropic “won”—it’s whether this decision creates enough cultural and legal scaffolding that other companies will feel emboldened to draw similar lines without getting crushed.
The Bigger Picture: AI Companies as Moral Actors
For the past five years, the AI safety movement has been mostly academic. People published papers. Conferences happened. Ethicists wrung their hands. But enforcement? Consequences? Those were abstract.
AnthropIC’s lawsuit—and this injunction—marks the moment when refusing to participate in potentially harmful applications becomes legally defensible. That’s a platform shift. Not because one company won one round, but because the legal system just validated the principle that an AI company can say no and expect to be protected for it.
It also exposes how much the Pentagon (and probably other agencies) were relying on the assumption that AI companies would eventually fold. Turns out, you can push back. You can sue. You can win, at least in the beginning.
The real test comes next. Does this ruling embolden other AI labs? Do they start writing similar clauses into their contracts? Or do they quietly continue the old arrangement—building capable systems and outsourcing ethical responsibility to customers?
My guess? Some will use this as cover to do the right thing. Others will find workarounds. Welcome to the messy middle.
🧬 Related Insights
- Read more: Could You Be the AI Sheriff Europe Needs? Inside the EU AI Office’s Urgent Hiring Spree
- Read more: Europe’s AI Literacy List: Savior for SMEs or Consultant Goldmine?
Frequently Asked Questions
Why did the Pentagon try to blacklist Anthropic? The DOD designated Anthropic a “supply-chain risk” in apparent retaliation for the company’s refusal to remove contractual language prohibiting its AI models from being used for mass surveillance. The court found this retaliation likely violated Anthropic’s First Amendment rights.
Can the Pentagon still appeal this decision? Absolutely. This is a preliminary injunction, not a final ruling. The case continues, and the DOD will likely appeal to higher courts. The injunction just prevents the blacklist from staying in place while that process unfolds.
Will other AI companies use this ruling to refuse government contracts? Possibly. The ruling creates legal cover for companies to include similar ethical restrictions in their contracts without facing immediate economic retaliation. However, whether companies want to restrict government use is a separate question—many have financial incentives to say yes.
