Anthropic: When a dry drunk demands your number and promises not to use it

On Friday, Trump barred an American AI developer, Anthropic, from doing further business with the federal government, and barred all contractors from doing business with Anthropic, an extreme punishment typically reserved for adversarial countries. Anthropic’s crime? Refusal to let the Department of Defense use its AI system, Claude, for surveilling American citizens or in autonomous weaponry that removes humans from decisions to kill.

Defense Secretary Pete Hegseth- the man who group texted attack plans to a reporter, wanted to punish an astronaut for stating the law, then shot party balloons with potent lasers despite FAA warnings that the lasers could blind pilots while they were in the sky with passengers, demanded that Anthropic let him use its AI system without contractual restrictions. When Anthropic said no, Trump blacklisted them.

It’s hard to say what’s more appalling- that the Trump administration is building tools for mass public surveillance like China’s, or that an undisciplined dry drunk like Hegseth has access to lethal toys.

Keeping up with China… in the worst way

Trump has said he wants to keep up with China through ‘global technological dominance’ and the ‘widespread use of AI.’ China’s authoritarian government uses one of the most advanced public surveillance systems in the world, collecting extensive facial recognition, biometric data, and personal profiles from private citizens against their wishes.

China captures these data from citizens’ faces, conversations, social media posts, phones and other devices while people stand at crosswalks, ride the bus, and go to the store, then feeds the data into an AI database used for oppression: for law enforcement, “monitoring social behavior,” and controlling access to services.

China’s system is similar to what Trump oligarch-supporting Peter Thiel’s Palantir is building, namely, a high-level data integration platform that will enable U.S. law enforcement, ICE, the IRS, DHS, DOJ, the military, and any other rogue agency Trump wants to weaponize to collect facial recognition, license plate readers, and other biometric data for mass surveillance.

Poor Hegseth, nobody believes him

There were clauses in Anthropic’s existing contract with the DOD that prevented Claude from being used for either mass surveillance of Americans or autonomous weaponry. While Anthropic had previously integrated Claude into some classified military networks, its $200 million contract with the federal government expressly prohibited using it for mass surveillance of Americans as well as autonomous weaponry, or ‘killer robots’ that can identify, select, and kill targets without a human in the decision-making loop.

These were the contractual restrictions Hegseth’s DOD demanded be removed. But Anthropic wasn’t having it.

Just before Trump blacklisted them, Anthropic’s CEO, Dario Amodei said the company could not, “in good conscience” agree to the Pentagon’s request. Amodei has expressed concern that Claude could be used for mass surveillance by automatically assembling “scattered, individually innocuous data into a comprehensive picture of any person’s life,” which seems to be exactly what Trump is trying to do.

In a series of angry social media posts, Undersecretary of Defense Emil Michael accused Anthropic of “lying” about using Claude for mass surveillance because the Dept. of Defense “doesn’t do mass surveillance as that is already illegal.”

Apparently the DOD does do comedy, because the suggestion that this regime will follow the law is a joke. Forget about the hundreds of court orders Trump has already violated. How many people have been murdered off the coast of Venezuela with zero legal justifications given? Claiming without evidence that we’re in an “armed conflict” with “narco-terrorists” is not a legal justification; it’s a dictator’s ‘shoot now, ask questions never’ strategy for breaking the law.

What can the AI at issue do?

Most Americans are blissfully unaware of how the emerging AI landscape could change their lives, and not for the better. Since I’m no AI expert, I asked Google AI to explain in simple terms how Anthropic’s Claude, if left to Hegseth’s command, could be used to spy on Americans. Here’s how AI described Claude’s functional capacity, verbatim:

· “Mass Data Synthesis (Sorting Huge Amounts of Info): Imagine a super-fast robot reading billions of text messages, emails, and internet posts all at once. It looks for “moods” (like who is angry or unhappy) and makes a map of where those people live.

· Intelligence Dossiers (Digital Secret Files): Using smart computer programs to read thousands of pages of documents about one person instantly. It acts like a digital detective, putting together a secret file on someone’s whole life.

· Automated Tracking (Digital Footprints): Looking at where people drive, what websites they visit, and who they talk to. This combines records to draw a map of where someone goes, like cameras on streets tracking cars.

· Law Enforcement Support (Police Tech Tools): Companies like Palantir create software for the police. This software combines information from cameras, bank records, and phone calls to track suspects and help police find them quickly.”

The Anthropic dispute and blacklisting has put Silicon valley on edge; if Trump and Hegseth can change the terms of AI contracts after the fact, what does a contract with this federal government even mean?

The regime’s dishonesty isn’t helping. Before Trump blacklisted Anthropic, Pentagon officials said they had “no interest” in using the illegal surveillance tools outlined above, while simultaneously seeking unfettered access to them.

Color me, and anyone with half a brain, skeptical.

Sabrina Haake is a columnist and 25+ year federal trial attorney specializing in 1^st and 14^th A defense. Her Substack, The Haake Take, is free.