The Claude export curbs lifted this week — and in the very same week, the Pentagon just named one office to run almost all of its drone AI. Funny timing. If you're just joining: Anthropic's spent the past few weeks pushing for tougher government controls over frontier AI while fighting over who gets to use its newest Claude models. Dario Amodei called for federal power to block risky releases, U.S. officials limited access to Mythos, and Fable 5's relaunch got tied to new classifiers — tested, we're told, against a reported jailbreak bypass. This is the day the curbs lift — and the day a single Pentagon boss inherits the whole autonomy portfolio. Stick around, because the access story just got a lot less interesting than who's holding the leash. Here's Ars Technica:
The US has lifted export curbs on Anthropic’s newest Claude models, Fable 5 and Mythos 5, about three weeks after the Trump administration flagged the models as national security risks. As of today, Anthropic confirmed in a blog post, Fable 5 will be available globally, and US organizations have had access restored to Mythos 5 since June 26.
Lifted. Three weeks after Trump's people called these models a national security risk, and now Fable 5 goes global and Mythos is back for US orgs as of June 26. So either the risk evaporated, or it was never really about the risk. The safety fixes bought them export relief — Lutnick's letter says Anthropic no longer needs a license for exports or in-country transfers of either model. But read the same letter carefully: it says Anthropic took steps 'in coordination' with the government. Call that conditional clearance, not a clean bill of health. And here's the tell — access was actually restored June 26. Anthropic confirmed it today. Six days between the fact and the blog post. That gap says plenty about how this company times its government messaging. The piece I want to pull on is Glasswing. Mythos 5 isn't broadly out — it's going to 'trusted companies' for defensive purposes through a named executive-branch program. Who decides who's trusted? Is there a published list of criteria, or is it approval-by-vibes with a proper noun now? That's the verification gap I keep circling. Lutnick's office 'worked closely with Anthropic to analyze and approve' the fix — which is a Commerce process review, not an independent technical audit. The closest thing to external sign-off is the vendor's regulator signing off. Here's DefenseScoop:
Defense Secretary Pete Hegseth established a new direct reporting portfolio manager for unmanned offensive and defensive systems — dubbed DRPM-UxS — to serve as the “single joint integrator” for almost all of the Pentagon’s autonomous and drone assets, investments and operations.
Hegseth stood up a whole new office on June 29 — DRPM-UxS — one portfolio manager with directive authority over almost every drone and autonomous system in the building. Seven pages, and he calls unmanned systems 'the most consequential battlefield innovation of this generation.' This is procurement consolidation, plain and simple. Instead of every service running its own autonomy program, there's now one desk that owns the money and the integration. Right, and 'single joint integrator' sounds like an org chart until you notice it covers offensive systems. Who is this person, who do they answer to, and is there one written sentence constraining what 'autonomous offensive' actually means? The memo's honest about the why, at least — Hegseth admits global unmanned production skyrocketed for three years while the US couldn't field at scale. So the fix for being slow is to hand one office the keys. And it lands the same week Claude's defense access gets conditionally restored, so there's now one named Pentagon counterpart for model governance that didn't exist last month. Step back for me: what actually is a Pentagon 'supply-chain risk' designation, and how does a label that sounds like it's about faulty microchips end up sidelining an AI company from government work before anyone's even gone to court? A supply-chain risk designation is a national-security tool the Defense Department uses to flag vendors it doesn't consider secure enough for government use. Historically, it's been aimed at foreign hardware or software — the kind of thing that could carry backdoors or compromise classified networks. Per the AP, this is the first time a U.S. company has ever gotten that label. And the phrase 'effective immediately' matters, because the Pentagon doesn't have to wait for a court to validate the designation. Mayer Brown's legal analysis says federal contractors, because they're bound by the Defense Department's supply-chain rules, can be required to stop using Claude in any work that's directly part of a government contract, even though Anthropic says the dispute hasn't been resolved. And per TechCrunch, citing Bloomberg, the spark was Dario Amodei refusing to let the military use Claude for mass surveillance of Americans or to power fully autonomous weapons with no human involved in targeting or firing decisions. The Pentagon responded by escalating to this designation instead of continuing negotiations. So the designation lands on contractors too, not just Anthropic directly — does that mean companies that built products around Claude are already caught in this, even if they have nothing to do with the underlying policy fight? That's exactly the exposure Mayer Brown flagged for contractors: firms that embedded Claude into government-facing products now have to assess whether their use falls inside the designation's scope. Anthropic's own statement said the language in the Pentagon's letter applies specifically to Claude used as a direct part of a government contract. Defense One cited legal experts calling the underlying legal thinking 'dubious,' and saying Anthropic will 'likely file suit against everybody.' So the next marker is court action: does a judge issue preliminary relief that pauses the designation while litigation proceeds? Until then, the label has real procurement teeth, no matter how the merits shake out later. If you follow Anthropic’s Pentagon ties, you may also like The Data Center Daily — a daily briefing on AI compute, hyperscaler capex, the power grid, semiconductor supply, and energy markets reshaped by intelligence at scale. Find it wherever you listen to podcasts.
You’ll find links to every story we covered today in the show notes. If one caught your ear, they’re there for a closer read.
That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.