The President just ordered every U.S. agency to stop using Anthropic — and on the same day, Dario Amodei is on the record asking for government power to block dangerous AI. You couldn't write the irony any cleaner if you tried. This is Anthropic Pentagon Watch. For a week, we framed this as a Pentagon procurement spat — today it's a government-wide freeze with a presidential signature on it. Government-wide. Not a Pentagon label, not a contract clause — all agencies, ordered to stop. They called the bluff, and yes, it was real. And it sits right next to the June 2 executive order telling fifteen agencies to accelerate AI adoption. Two executive instruments, opposite directions. Let's start there. Here's the legal edge: Goodwin told us not to wait for formal acquisition paperwork to give the ban teeth. A presidential directive to all agencies may be the paperwork. Which means the live question is what happens to incumbents. The NSA's Mythos contract, the engineers embedded running offensive cyber — do they rip it out, issue a waiver, or just stop expanding? The June 2 order has a 30-day review window. Does it actually say? That's the part nobody's confirmed yet — what happens to projects already running Claude. The order says stop. It doesn't tell you how you unwind something that's already inside the building. Here's Europe Says:
WASHINGTON (AP) — The Trump administration on Friday ordered all U.S. agencies to stop using Anthropic’s artificial intelligence technology and imposed other major penalties, escalating an unusually public clash between the government and the company over AI safety.
Make it official — Trump ordered every U.S. agency to stop using Anthropic. Not just the Pentagon, not just defense contracting. Government-wide. The quote on the page is the President saying, 'We don't need it, we don't want it, and will not do business with them again.' That's a procurement freeze written in caps lock. And Hegseth slaps a 'supply chain risk' label on them — the tag you stamp on a foreign adversary. He just put an American AI lab in the same bucket as a hostile state. For three episodes, we asked whether the blacklist was a bluff. It wasn't. The White House went around the Pentagon entirely and cut Anthropic off at the agency level. The trigger should be clear — Amodei wouldn't strip the guardrails limiting military use by Friday's deadline. He declined, so they detonated the whole relationship. Crypto Briefing, with Estefano Gomez:
Anthropic CEO Dario Amodei said governments should be able to block artificial intelligence companies from releasing powerful new models if outside testing shows they pose unacceptable risks. In a Wednesday essay, Amodei argued that advanced AI systems should face mandatory third party evaluations before deployment, with reviewers assessing risks across areas such as cybersecurity, biological threats, and other misuse scenarios.
Dario Amodei put out an essay Wednesday saying governments should have the power to block dangerous AI models before release — mandatory third-party testing, federal veto, the works. And then days later, that same government ordered every agency to stop using his company's tech. So he's asking for a veto power that is — right now — being aimed at him. You want to find the punchline of this week, that's it. The new piece here is the ask. He compared frontier AI to cars, airplanes, drugs — strict safety rules before public release. He's talking binding regulation, beyond the voluntary transparency pledges everybody's been hiding behind. Right, but track who holds the veto. He says the government should have authority to block. Which government? The one that just blacklisted him, or some clean regulator that doesn't exist yet? Because those are very different deals. And he's careful — the essay never says who runs the third-party evals. Convenient gap for a CEO who's currently litigating against the people who'd appoint the reviewers. Global Policy Watch writes:
While the Order establishes a process for government and industry collaboration on AI security, the voluntary nature of this process and its specific focus on cybersecurity reflect the Administration’s stated commitment of avoiding burdensome federal regulation of AI technology while at the same time protecting critical infrastructure and national security information systems against new cyber threats posed by “advanced AI.”
Here's the document everyone's stepping around. June 2, an executive order titled 'Promoting Advanced Artificial Intelligence Innovation and Security' — and the security framework in it is voluntary. No legal requirement for any lab to submit a model for review. Voluntary. So the same week Trump orders every agency to stop touching Anthropic, the actual security framework is a polite invitation nobody has to RSVP to. And it gets better — the Order doesn't define 'advanced AI' or 'covered frontier models.' So the whole thing rests on agency implementation that hasn't happened yet. Undefined terms, voluntary submission, and a blanket ban running in parallel. The carrot is optional and the stick is presidential. That tells you which one they actually mean. Two executive instruments pointing opposite directions — one says collaborate on AI security, the other says freeze Anthropic out of every agency. Contracting officers get to reconcile that on their own time. Step back for me: Anthropic is in a standoff with the Pentagon over how Claude gets used, but Dario Amodei is also out there saying governments should be able to block dangerous AI — so which is it? Is the company for government control of AI or against it? Those are two different kinds of government power. Amodei's been calling for binding federal rules on frontier AI: in an essay this month, he said transparency requirements alone aren't enough, and he told ABC News the government should have a narrow power to block deployment of unsafe technology. That's oversight of what gets built and released. The Pentagon fight is about how the product gets used once the government buys it — specifically, the military pressing Anthropic to drop what the 80,000 Hours podcast describes as its ban on AI-only kill decisions and mass domestic surveillance. And remember, in Amodei's own February statement, Anthropic said it was the first frontier AI company to deploy models on U.S. classified networks and at the National Laboratories. So the company does work with government. It's drawing lines around specific uses, no matter who's asking. Critics like Marc Andreessen call that hypocrisy — you can't be pro-regulation and then resist government control. But that charge blurs two separate fights. So when Anthropic says 'regulate AI,' they mean safety rules applied to everyone, including themselves — but what the Pentagon wanted was for Anthropic specifically to remove its own internal limits on how the military uses the model? That's the core of it. Government rulemaking is one lane. A procurement deal that overrides a vendor's product constraints is another — here, constraints on autonomous weapons decisions and mass surveillance. The Pentagon answered by designating Anthropic a supply-chain risk. That's the piece to keep watching: it has real procurement consequences, and however courts or contracting officers treat it will likely set a precedent for every AI company that tries to sell to the government with conditions attached. If you’re tracking AI power and accountability, try Musk v Altman Daily — a daily court-watch on Elon Musk’s trial against Sam Altman, OpenAI, and Microsoft, covering testimony, exhibits, and the AGI governance fight. Find it wherever you listen to podcasts.
You’ll find links to every story we covered today in the show notes. If something caught your ear, they’re there for a closer read.
That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.