Google Steps In After Anthropic Tells the Pentagon No

Anthropic made a public stand last month. The company updated its acceptable use policy to explicitly prohibit the Department of Defense from using its Claude models for “domestic mass surveillance” and “autonomous weapons.” No ambiguity, no fine print. Just a clear line in the sand.

Google, meanwhile, just signed a new contract with the Pentagon.

Let me be clear about what happened: Anthropic didn’t just quietly decline a contract. They put out a policy change that directly called out the DoD by name, stating their models couldn’t be used for those specific purposes. It was a deliberate, public positioning. And now Google has stepped in to fill the gap.

The new deal reportedly expands the Pentagon’s access to Google’s AI capabilities, including its Gemini models and cloud infrastructure. Google has been working with the DoD for years through its Google Cloud division, but this agreement goes further. The Pentagon will now have broader access to Google’s AI tools for what they describe as “national security missions.”

I’ve seen this play out before. Google’s “Don’t Be Evil” era ended a while ago, but Project Maven in 2018 was a real turning point. That was the drone imagery analysis contract that sparked internal protests and ultimately led to Google not renewing it. Now they’re back in a bigger way.

What’s interesting is how differently these two companies view the same ethical questions. Anthropic was founded by former OpenAI researchers who wanted to build safe AI. They’ve been vocal about responsible deployment. Google has its own AI principles, but they’ve proven flexible when the money is right.

Anthropic’s move wasn’t just about optics. Their policy specifically prohibits “using Claude to control or direct autonomous weapons systems” and “mass surveillance of domestic populations.” Those are concrete restrictions that would have made a DoD partnership legally complicated. Google’s principles, by contrast, are more about “socially beneficial” applications and avoiding “harm.” That’s vague enough to accommodate a lot of military work.

Here’s the uncomfortable reality these companies face: the DoD has money and wants AI. If one company says no, another will say yes. The Pentagon isn’t going to stop pursuing AI capabilities just because Anthropic has principles. They’ll find someone else.

Google’s contract is reportedly worth hundreds of millions of dollars. That’s not pocket change, even for a company with Google’s revenue. And the DoD gets access to some of the most capable AI systems currently available, including Gemini’s multimodal capabilities and Vertex AI for custom model deployment.

I don’t think this is a simple good-versus-evil story. Google will argue they’re helping defend the country. Anthropic will argue they’re drawing necessary ethical boundaries. Both positions have merit. What bothers me is the lack of transparency. We don’t know what specific projects this contract covers. “National security missions” could mean anything from logistics optimization to weapons targeting.

The timing is notable too. This comes as the Biden administration pushes for more AI regulation and as Congress debates the National AI Commission Act. The military AI contracts are happening regardless of what the policy debates produce.

Anthropic’s refusal sets a precedent. Other AI companies will have to decide where they stand. But Google’s willingness to step in shows that ethical posturing only goes so far when there’s a willing buyer and a willing seller. The Pentagon will get its AI. The only question is which company provides it.