The Trump administration has decided that Anthropic's most powerful AI models are too dangerous for foreign governments to touch, which would be a perfectly coherent national security position if the Pentagon hadn't already decided those same models are too dangerous for the U.S. government to touch. Howard Lutnick sent a letter to Anthropic CEO Dario Amodei on Friday formally notifying him that the Mythos 5 and Fable 5 models are now subject to a Commerce Department export licensing regime. Congratulations, Dario. You built something so threatening that America won't let anyone use it, including America.
What Lutnick's Letter Actually Says
According to Axios, Commerce Secretary Howard Lutnick sent the letter to Anthropic CEO Dario Amodei on Friday, formally placing the Mythos 5 and Fable 5 models under a licensing regime that blocks foreign governments, companies, and individuals from accessing them without explicit U.S. government approval. This is the Commerce Department's tool of choice when it wants to treat a technology like a weapons system. You want it? You have to ask Washington first.
The framing from the administration is straightforward: cutting-edge AI is now a national security asset, the same way advanced semiconductors, certain drone components, and high-grade encryption technology are national security assets. The logic isn't crazy on its face. If your AI model is genuinely that powerful, keeping it out of adversaries' hands is a reasonable thing to want.
The Part Where It Gets Weird
Here's the part that deserves more attention than it's getting. As Axios reports, Anthropic simultaneously finds itself on a Pentagon blacklist deeming its models too dangerous for the U.S. government's own use. Read that again slowly. The federal government has decided these AI systems are too sensitive and dangerous for foreign powers to access, and also too sensitive and dangerous for the federal government to access.
That is not a coherent policy. That is two different bureaucracies pulling in two different directions and nobody in the White House apparently noticing, or caring, that the positions are in direct tension with each other. The Pentagon is essentially saying: we don't trust this thing. The Commerce Department is simultaneously saying: we must protect this thing from the world. Who exactly is this protection serving?
It would be like the FDA banning a drug from export for being too effective while the VA simultaneously bans it from military hospitals for being too dangerous. At some point someone has to get in a room and decide what they actually think.
Anthropic's Genuinely Awkward Position
Anthropic has spent the last three years very carefully cultivating an image as the responsible AI company, the one run by former OpenAI researchers who left because they were worried about safety, the one that published detailed model cards and constitutional AI papers and generally tried to be the adult in a room full of people racing toward the cliff. That positioning has now produced a genuinely surreal outcome: their caution and their capability have combined to land them in a regulatory vise.
They are blacklisted by the Pentagon. They are under export controls at Commerce. They are, in the parlance of Washington bureaucracy, a problem to be managed rather than an asset to be deployed. The safety-first AI company has been regulated like a weapons manufacturer. There is a dark irony in there somewhere that Dario Amodei is probably not finding very funny right now.
Anthropol has not yet made a public statement in response to the Lutnick letter, according to Axios reporting. What exactly would you even say?
Washington's Escalating AI Crackdown
This move fits a broader pattern that Axios correctly identifies as an escalation. Washington has been slowly, then quickly, reframing advanced AI as a national security domain rather than a commercial technology sector. Export controls on AI chips. Entity list additions targeting Chinese AI labs. And now direct licensing requirements on specific model versions from specific American companies.
The Trump administration did not invent this trend. The Biden administration laid substantial groundwork with chip export controls and the AI executive order. But the current administration has leaned into the national security framing harder and faster, and it is now producing situations like this one, where a San Francisco AI startup is being treated with the same regulatory seriousness as a defense contractor building submarine components.
The question nobody in Washington seems to want to answer out loud is where this ends. Every six months the frontier models get more capable. Are we going to slap export controls on every major model release indefinitely? What does a sustainable policy actually look like?
What This Means for the AI Industry
The practical effect for Anthropic's business is significant. Foreign companies that have been using Claude, Anthropic's commercially available model family, through the API will now have to determine whether Mythos 5 and Fable 5 fall under the licensing requirement and what that means for their operations. Compliance costs. Legal uncertainty. The general headache of operating at the intersection of a fast-moving technology and a slow-moving regulatory apparatus.
For the broader industry, this is a signal. If Anthropic, which has been meticulously careful about safety and government relations, can end up simultaneously Pentagon-blacklisted and Commerce-controlled, no frontier AI lab is safe from this kind of regulatory whiplash. OpenAI, Google DeepMind, Meta AI, all of them are watching this unfold and quietly calling their regulatory affairs teams.
The Dingo Take
Let's be honest about what the Trump administration has done here. They have constructed a policy position that manages to be maximally restrictive while being minimally coherent. Anthropic's best AI is too dangerous for America's enemies. Also too dangerous for America's military. The logical conclusion of this framework, if you follow it out, is that the technology should not exist at all, but nobody is saying that because the administration is also simultaneously claiming American AI dominance as a geopolitical victory. You cannot have all three of those positions at once.
The deeper problem is that Washington is making up AI policy in real time, agency by agency, letter by letter, with no overarching framework that anyone has actually committed to. The Pentagon blacklist and the Commerce licensing regime appear to reflect entirely different threat assessments of the same technology, and the fact that nobody caught this contradiction before Lutnick sent his letter suggests the interagency coordination on AI policy is basically nonexistent. That is a genuinely scary thing when the stakes are this high.
Anthropological could be forgiven for feeling like they did everything right and got punished for it anyway. They built something powerful. They were transparent about it. They flagged the risks themselves, repeatedly, publicly, in writing. And the government's response has been to treat them like a problem to be contained. If that's the lesson the rest of Silicon Valley takes from this, the age of voluntary AI safety cooperation with Washington is going to be very short indeed.
