Google Steps In After Anthropic Tells the Pentagon No

Anthropic made a very public decision last week: it wouldn’t let the Department of Defense use its Claude models for domestic mass surveillance or autonomous weapons. The company framed it as a matter of principle, drawing a firm line around what it considers unacceptable military use.

Now Google has stepped in. The company signed a new contract with the Pentagon that expands the military’s access to its AI tools. The timing is interesting — and not exactly subtle.

Let me be clear: this isn’t a secret deal. It’s a formal expansion of Google’s existing work with the DoD, which includes cloud infrastructure and AI capabilities through its Google Cloud division. The Pentagon gets more access to Google’s models for a range of applications, and Google gets a multi-million dollar contract and a seat at the table for defense AI.

The contrast with Anthropic is sharp. Anthropic’s refusal wasn’t just about ethics — it was a brand move. The company has positioned itself as the “responsible AI” player, and walking away from Pentagon money reinforces that image. But it also leaves money on the table, and Google is happy to pick it up.

I don’t think this makes Google evil or Anthropic saintly. The reality is messier. Google has its own ethical AI principles, but they’re flexible enough to accommodate military contracts. The company learned from the Maven project backlash in 2018, when employees protested the Pentagon drone imagery contract. This time around, the internal pushback seems quieter, or at least less public.

The Pentagon, for its part, doesn’t care about the moral theater. It needs AI capabilities, and if one vendor says no, it moves to the next. Anthropic’s refusal is a speed bump, not a wall. The DoD has been working with Google, Microsoft, Amazon, and Palantir on AI for years. Losing one supplier barely registers.

What bothers me is the lack of transparency around what “expanded access” actually means. Google’s announcement is vague — it mentions “advanced AI capabilities” and “enhanced collaboration” but doesn’t specify which models, what data the Pentagon can feed into them, or what safeguards exist. Given Google’s track record, I’d like to see more detail before cheering or condemning this.

Anthropic’s stance is notable, but it’s also a luxury. The company can afford to turn down military contracts because it’s not dependent on them for revenue. Google, with its massive cloud business and government sales team, operates differently. The Pentagon contract is a drop in the bucket for Google’s bottom line, but it opens doors for bigger deals down the line.

I suspect we’ll see more of this pattern: smaller AI companies take principled stands, while the big platforms quietly absorb the business. It’s not a conspiracy — it’s just how incentives work. Anthropic gets good press and a clear conscience. Google gets a contract and a foothold. The Pentagon gets its AI.

The real question is whether this arrangement leads to better oversight or just more opacity. I’m not optimistic. Military AI contracts are classified, ethical guidelines are voluntary, and the public rarely gets to see what the models are actually doing. Anthropic’s refusal is a headline, but Google’s expansion is the reality.

If you care about where AI is headed, pay attention to the quiet deals, not the loud refusals.