Luxury unsolicited moral philosophy, delivered with the conviction of a defense contractor’s press release: Google has agreed to give the U.S. Department of Defense access to its AI systems on classified networks for “all lawful uses.” This makes Google the third major AI company — after OpenAI and xAI — to sign such a deal with the Pentagon. The deal happened because Anthropic, the company that literally has “safety” as a personality trait, told the Pentagon no.

And then the Pentagon called Anthropic a supply-chain risk. A designation normally reserved for foreign adversaries. Because nothing says “healthy democracy” like labeling your most safety-conscious AI lab with the same tag you put on Huawei.

🤚 The Open-Palm Recap

Here’s what happened, in order of escalating discomfort:

1. The Pentagon asked Anthropic for unrestricted access to its AI on classified networks.
2. Anthropic said yes, but with conditions: enforceable guardrails preventing use in domestic mass surveillance and autonomous weapons systems. Reasonable stuff, you’d think.
3. The Pentagon said no to the conditions. Apparently, “we promise not to use it for autonomous killing machines” was too much paperwork.
4. The Pentagon designated Anthropic a “supply-chain risk” — the bureaucratic equivalent of keying someone’s car because they wouldn’t lend you their house keys.
5. Anthropic sued. A federal judge granted an injunction in March 2026. Litigation continues.
6. Google, OpenAI, and xAI all signed comparable deals without the fuss.

And here we are. The one company that tried to set boundaries is in court, and the three that didn’t are cashing checks.

👐 The Two-Handed Irony

This is where the story gets truly exquisite. Google — the company now providing unrestricted AI to the Pentagon — is simultaneously one of Anthropic’s biggest investors, having recently committed a \$40 billion investment in cash and compute. That’s right. Google is funding the company the Pentagon is punishing for having ethics, while also being the company the Pentagon hired because Anthropic had ethics.

It’s like paying your neighbor’s mortgage while also dating their ex. Technically not illegal. Spiritually unconscionable. Absolutely happening.

Google’s contract reportedly includes language saying it “doesn’t intend” for its AI to be used for mass surveillance or autonomous weapons. The Wall Street Journal reports this clause is essentially non-binding. It’s the national security equivalent of a “please don’t feed the bears” sign — technically there, functionally decorative.

👑 The Gold-Leaf Employee Revolt

950 Google employees have signed an open letter opposing the deal and urging leadership to follow Anthropic’s example instead. Nine hundred and fifty. That’s not a petition — that’s a small village’s worth of engineers who presumably have very strong opinions and very active Slack channels.

This echoes the 2018 Project Maven protests, when Google employees successfully pressured the company to drop a Pentagon drone imagery contract. But the stakes are incomparably higher now. In 2018, AI was a novelty that could identify objects in drone footage. In 2026, AI can reason, plan, and operate autonomously. The difference between then and now is the difference between lending someone binoculars and lending them a brain.

Google’s leadership has so far responded to the open letter with the corporate equivalent of “we hear you and we value your perspective” — which, translated from executive-speak, means “we read it and we’re doing it anyway.”

🌿 The Gentle Awakening

Let’s sit with the uncomfortable geometry of this situation:

• Anthropic tried to set enforceable limits on military AI use and was punished by the government
• Google signed without enforceable limits and was rewarded with a contract
• The market lesson is unmistakable: safety has a price, and the Pentagon isn’t paying it

The supply-chain risk designation is particularly chilling. If the U.S. government can label a domestic AI company — one that wanted to work with the Pentagon, just with guardrails — as a supply-chain risk for having conditions, then the message to every AI lab is clear: cooperate unconditionally, or be treated like an adversary.

This is the kind of precedent that makes “responsible AI” less of a corporate value and more of a business liability. And somewhere in a conference room, Anthropic’s lawyers are preparing arguments while Google’s shareholders are preparing champagne.

Different rooms. Same building. Same industry. Wildly different conclusions about what the word “responsible” means when there’s a defense budget on the table.

Google’s AI was unavailable for comment on the ethics of the deal, but it did helpfully generate a PowerPoint presentation titled “Synergizing Defense Capabilities Through Responsible Innovation” in under four seconds.