The ban's live, and nobody on the contractor side agrees on what it actually makes you sign. If you're just joining: Anthropic and the Pentagon have been locked in a procurement fight over whether Claude can be used for classified military work while Anthropic keeps limits on lethal autonomous weapons and mass surveillance. At the same time, federal adoption kept moving — CISA reportedly used Mythos for government code audits, even as the supply-chain-risk designation left agencies and contractors sitting in legal uncertainty. This is Pentagon Watch. Today — the ban's real, the export controls are already lifted, and someone finally mapped the certification maze contractors are staring at Monday. We're reading the receipts now. Start with the paperwork nobody wanted to open. Mayer Brown put names on it. We'll keep tracking Anthropic-Pentagon supply-chain-risk fight — follow the show so the next update finds you. Ryan Frazee, Adam S. Hickey, and John Prairie, writing in Mondaq:
The Department of War (“DoW”) is now moving out on enforcement of its March 2026 designation of Anthropic as a “supply chain risk,” the first such designation ever applied to an American company. Contractors across the defense-industrial base are receiving agency and prime-contractor requests to identify, remove, and certify non-use of Anthropic products and services, including Claude, and those requests are arriving quickly and from multiple directions at once.
The ban is live. The March supply-chain-risk designation on Anthropic — the first ever applied to an American company — is now generating actual paperwork, and Mayer Brown just published the contractor-side breakdown. And here's the mess: multiple agencies, multiple primes, all sending different certification requests. Different scopes, different deadlines, different attestations about whether you touch Claude at all. A patchwork means somebody wrote each patch. With Frazee, Hickey, and Prairie on the Mayer Brown byline, the billable hours are already flowing, and the primes are gonna eat the compliance cost while the little subs get squeezed out of the pipeline. That's the piece that lands for me. When a firm that reps major funds and financial institutions publishes a public how-to-certify explainer, counsel is telling clients what liability they're holding as of Monday morning. So contractors get a compliance maze for touching an American lab that said no to autonomous weapons. Meanwhile, nobody's sending certification letters about the PLA-linked firms on the other side of the ledger. Funny how the paper fence only faces one direction. From Tamzid at Bright Defense:
Anthropic restored global access to Claude Fable 5 on July 1, 2026, ending an 18-day suspension triggered by U.S. export controls over a reported jailbreak discovered during Amazon security testing. Although the restrictions were lifted, the incident exposed unresolved questions about government oversight of frontier AI models, export-control authority, enterprise AI governance, and the responsibilities of organizations deploying advanced AI systems.
Here's the fact that reframes the whole export-control panic: the jailbreak that got Fable 5 yanked for 18 days? Amazon's security team found it. Not Anthropic's red team, not a DoD audit — Amazon. And access came back July 1. So the government slapped an emergency export control on the exact model it called a national-security risk, then quietly lifted it eighteen days later once the fix was in. So the trigger for a global shutdown was a finding from Anthropic's own cloud partner. Follow the contract trail on that one — Amazon's running the infrastructure, and Amazon surfaced the hole. And Bright Defense flags the wrinkle — Mythos 5 is still restricted. So whatever satisfied the government on Fable 5, the sibling model didn't clear it. From Yale Journal on Regulation:
On April 7, 2026, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened the chief executives of five systemically important banks to brief them privately on Claude Mythos Preview, a frontier AI model that Anthropic declined to release to the public. Anthropic made this decision upon finding in its internal testing that the model could discover latent software vulnerabilities at a scale and speed beyond any existing cybersecurity tool or human effort.
Now the meeting nobody was supposed to see: April 7, Bessent and Powell pulled five bank CEOs into a room to brief them on Claude Mythos Preview. And what came out of it? No rules, no written record, no public statement. Read that back. The Yale Journal piece takes that closed-door briefing and tries to build a procedure around it — explicitly modeled on the Treasury-Fed convening of systemically important banks. That's a genuinely consequential move. Because if you designate Anthropic as systemically important the way you'd designate a bank, that stops being rhetoric. It becomes a legal-governance status with obligations and leverage attached — and it cuts both ways on the concentration worry. Right, but back up — we just walked through the Department of War ban, where contractors are drowning in a certification patchwork. Same lab. One arm of the government is blacklisting Anthropic; another is convening bank CEOs around its restricted model. Whose framework wins? Got a tip, a story idea, or a correction for us? Send it our way at anthropicpentagonwatch at lantern podcasts dot com. We read every note, and it helps sharpen the briefing.
Links to every story we covered today are in the show notes. If something caught your ear, that's where to go for a closer read.
That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.