Google steps in after Anthropic tells the Pentagon ‘no’ on surveillance and weapons

Here’s something that landed with a bit of a thud this week: Anthropic told the Department of Defense they couldn’t use Claude for domestic mass surveillance or autonomous weapons. Google, on the other hand, just signed a new contract with the Pentagon.

Let me be clear — Anthropic didn’t just say “no thanks” quietly. They made a principled stand. Their acceptable use policy explicitly bans using their models for things like warrantless surveillance at scale or weapon systems that can decide to kill without human input. When the DoD came knocking with certain requests, Anthropic held the line.

And Google? Google saw an opening.

The new agreement gives the Pentagon expanded access to Google’s AI models, including Gemini, for a range of military applications. Google’s official line is that they’re “committed to supporting the U.S. government and military in a responsible manner,” which is corporate speak for “we’ll take the money.”

I don’t want to pretend this is simple. The Pentagon is a massive customer, and AI contracts with them are worth serious money. Google has been trying to rebuild trust with the military after the whole Project Maven fiasco back in 2018, when thousands of employees protested the company’s involvement in drone strike targeting. They walked that back then, but apparently the pendulum has swung again.

What’s interesting is how differently these two companies approach the same question. Anthropic was founded by former OpenAI people who explicitly wanted to build safer AI. Their whole brand is about responsible deployment. Saying no to certain military uses is consistent with that. Google, despite all the “Don’t Be Evil” nostalgia, has been drifting toward more aggressive government partnerships for years. This contract is just the latest evidence.

The specific capabilities the Pentagon gets access to aren’t fully public, but the contract reportedly covers a broad range of AI services — from logistics optimization to intelligence analysis to, yes, some of the stuff Anthropic refused to touch. Google’s safeguards are supposedly in place, but we all know how well corporate AI ethics boards hold up when there’s billions on the line.

This isn’t a new debate. The military-industrial complex has been buying tech since before Silicon Valley was a thing. What’s different now is the scale and the nature of the technology. AI isn’t just another tool — it’s decision-making infrastructure. Handing that over to the Pentagon means accepting that your models will be used in ways you might not control.

Anthropic took a stand. Google took a contract. I’m not saying one is evil and the other is pure — the reality is messier than that. But it’s worth paying attention to which companies draw lines and which ones don’t.