OpenAI Inks the Pentagon's 'Lawful' Pact—Anthropic Said No, Users Pulled the Rug

OpenAI just shook hands with the Pentagon to plug its AI into classified systems—a move that came right after the previous administration showed Anthropic the door for refusing the same terms. OpenAI is waving its "red lines" like a white flag at a bullfight: no mass domestic spying, no killer robots, no social credit scores. But the fine print? The contract permits use for 'all lawful purposes.' You know, that classic catch-all that's about as specific as a VC saying "we're building the future."

That magic phrase? It's the exact same one that made Anthropic grab its coat and leave the party.

OpenAI counters that its cloud-only setup is a built-in chastity belt against lethal autonomy. Fair enough. But if your AI is painting targets, sketching "patterns of life," or mapping out missions? Congrats, you're still on the payroll for the kill chain. Whether the final click is by a meatbag's finger or a server rack in Ashburn doesn't change the body count.

The surveillance clause swears off 'unconstrained' monitoring of U.S. persons—then immediately name-drops FISA, the Fourth Amendment, and EO 12333. For those keeping score at home, that's the legal equivalent of a "wink wink, nudge nudge"; it's the very framework that lets the NSA legally vacuum up American data the moment it sneezes outside a border.

Anthropic's take? The law is still buffering while AI is already streaming in 4K. You can't lean on 'lawful' as a defense when the three-letter agencies are casually shopping for 300 million private messages from a misconfigured S3 bucket.

Altman's copium? 'We'll bail if they ask us to do something illegal.' Note: not unethical. Illegal. He's putting his faith in technical safeguards and legacy statutes rather than hard contract bans. Anthropic wanted ironclad prohibitions. OpenAI wanted optionality and a backdoor for plausible deniability.

The users, in true degen fashion, conducted a rug pull of their own. Over 1.5 million joined the #QuitGPT exodus. Subscriptions were cancelled. Reddit threads melted down. Someone even tagged Anthropic's SF office with "Sellouts?" in spray paint. Katy Perry, in a peak crypto-twitter moment, just posted Claude's pricing page—the ultimate signal of distress.

The result? Anthropic's Claude app rocketed to the top of the Apple App Store charts. Free tier signups went parabolic, proving that in the attention economy, controversy is the best marketing airdrop.

Let's not get too sanctimonious, though. Anthropic's own cozy deals with Palantir and AWS? They're still piping Claude's brain to intelligence agencies. And yes, those tools have reportedly seen action in shadowy ops.

So the real gap here isn't moral high ground. It's a philosophical bet on whether 'lawful' is a protective shield or a giant, exploitable loophole waiting for a creative lawyer.

The crowd narrated the story. And, as it tends to do, the market priced in the sentiment.

Also, in unrelated news, scientists figured out how to turn milk protein into biodegradable plastic wrap. Because sometimes, the real disruption is figuring out how to keep your sandwich fresh without killing the planet.