Anthropic Sues the Pentagon

The fight over who controls AI in national security just went legal.

According to Reason, Anthropic refused to sign a Department of Defense contract unless the Pentagon explicitly promised NOT to use its Claude models for autonomous lethal weapons systems or domestic mass surveillance. The Pentagon's response? Label Anthropic a "supply chain threat."

That designation — if it had stuck — would have forced Amazon, Google, and Nvidia to sever business ties with one of America's most important AI companies. Anthropic sued. The case is now in active litigation alongside closed-door negotiations.

Defense Undersecretary Emil Michael went on CNBC's Squawk Box in March 2026 and said: "We realized we are dependent on this one provider who wants to insert their policy preferences in the middle of an operation."

Translation: the Pentagon is angry that a private company won't hand over its technology with zero ethical guardrails.

Anthropic isn't wrong to demand those limits. Autonomous weapons that make kill decisions without human oversight and AI systems that conduct mass surveillance of American citizens are not abstract concerns. They're the exact scenarios that should concern anyone who believes in individual liberty.

But the Pentagon isn't entirely wrong either. Dependency on a single AI provider that can unilaterally dictate battlefield rules is a real operational vulnerability.

OpenAI Went the Other Direction — and Got Protests for It

While Anthropic drew a line, OpenAI signed a major military access deal with the Pentagon, according to reporting cited by articles.emp0.

The company claims the agreement prohibits autonomous weapons and mass surveillance use, pointing to the 2023 Pentagon directive on autonomous weapons. OpenAI researcher Boaz Barak argued the company can embed ethical red lines — including no mass surveillance and no directing weapons without human involvement — directly into model behavior.

That triggered massive protests in London and elsewhere, with activists demanding stronger safety guarantees.

According to articles.emp0, models like Claude have already appeared in combat zones despite official bans. If the rules can't be enforced in practice, the rules don't matter.

Both sides of mainstream coverage are missing the same gap: the distance between what AI companies promise their military contracts prohibit and what actually happens when those models are deployed in classified settings nobody can audit.

Courts Just Ruled: Your AI Prompts Are Fair Game in Litigation

Federal Judge Thomas O. Farrish (D. Conn.) ruled last Monday in Conservation Law Foundation, Inc. v. Shell Oil Co. that AI prompts used by an expert witness are subject to compelled discovery under Rule 26(b), according to the Volokh Conspiracy.

Shell's lawyers demanded the AI prompts used by plaintiff's expert Dr. Naomi Oreskes in her analysis of Shell's documents. CLF argued the prompts were protected as "expert notes." Judge Farrish disagreed.

His logic: an expert's methodology is always fair game for discovery. If you used AI to sort, filter, or analyze documents, how you prompted that AI is part of your methodology.

This has massive downstream implications. Any expert witness — in environmental cases, patent disputes, medical malpractice, and beyond — who uses AI in their analysis now needs to document and preserve every prompt. Defense teams can challenge AI-assisted conclusions by demanding to see exactly what questions produced the answers.

The ruling changes how every courtroom will handle AI evidence going forward.

States Are Doing What Washington Won't

While the federal government fights over who controls AI warfare, states are quietly building the actual governance framework for AI surveillance on American soil.

According to Mobile Pro Systems, Colorado's SAFE Act (SB 26-071) and South Carolina legislation are moving forward in early 2026, alongside city-level frameworks in Austin and Seattle. These rules are codifying:

• What purposes surveillance tools can legally serve
• How long footage and data can be stored
• Who owns the data and controls access
• Which AI capabilities — facial recognition, vehicle tracking, predictive analytics — are permitted

This legislation determines whether Americans get surveilled without recourse. Law enforcement agencies that rely on mobile surveillance platforms and real-time crime centers are now operating in an environment where governance requirements are hardening fast.

The federal government has a 6-pillar cyber strategy and a White House AI framework. Neither one tells a sheriff's department in Colorado what to do with three years of license plate reader data. The states are filling that vacuum.

Bernie's Moratorium Idea Is Both Understandable and Dangerous

Sen. Bernie Sanders (I-VT) announced in 2026 that he's pushing legislation to impose a moratorium on new AI data center construction until national safeguards are in place, according to Reason.

His fear — that superintelligent AI could exceed human control and pose an existential threat — is not unfounded. Anthropic CEO Dario Amodei wrote in a 2026 essay that we are "so close to these models reaching the level of human intelligence" and that society hasn't grasped what's coming.

But a moratorium hands China a free runway. China is NOT pausing. Every month American AI development stalls is a month Beijing closes the gap.

The answer is to build the right rules while moving forward.

Where This Heads

The AI governance fight is no longer theoretical. It's in federal court, on state legislative floors, in military contracts, and in legal discovery battles over expert witness methodology.

Americans should care about all of it — because the rules being written right now will determine whether AI becomes a tool of individual empowerment or a weapon of government control. The Pentagon wants no limits. Sanders wants a full stop. Neither extreme serves the public.

The Anthropic lawsuit will set a precedent. The Colorado surveillance law will be copied. Judge Farrish's discovery ruling will change how every courtroom handles AI evidence. The decisions being made right now will shape the next decades.