A signed executive order, Anthropic walking into cyberthreat briefings inside agencies that were supposed to be off-limits, and Congress reaching for a pen — all three happened on the same day. This is Anthropic Pentagon Watch. Today we’ve got the EO, the 90-day pre-release window that got shaved to 30, and Capitol Hill drawing its own red lines — let’s take them one by one. The legal fight is still alive, but it’s not the only thing moving anymore. Which is either great news for Anthropic, or a sign their lawsuit is about to get buried under somebody else’s statute. Let’s start with what changed when Trump signed on June 2, because there’s a number in that text procurement teams are already staring at. This one's from The White House:
It is the policy of the United States to promote AI innovation and security by working collaboratively with the private sector to modernize government and private sector information systems and harden them against external threats; to protect American ingenuity and intellectual property from exploitation and theft by adversaries; and to cultivate America’s advanced AI-enabled capabilities.
The June 2 EO is signed, so now it’s real. And tucked inside is a 60-day clock: the executive branch has to build a classified benchmarking process to decide what counts as a covered frontier model. That definition is the whole procurement fight — it decides who gets into the pre-release access window and who stays outside it. And that window got cut back quietly — 90 days in the draft, 30 days in the final text. Somebody pushed that down, and if your lab isn’t already on classified networks, you’ve got thirty days to matter or get locked out of the tier that counts. Anthropic is still in active litigation with the Pentagon, and it’s not on the May 1 seven-company roster. So “thirty days of pre-release access” for Anthropic right now is a phrase that needs a lawyer to finish. Here's one from Hacker News (32 pts thread):
What seem like the most interesting bits, selectively pruned of legalese: >Within 60 days of the date of this order, the shall: >(a) develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a “covered frontier model” for the purposes of this order, sharing such assessments with AI developers and researchers as appropriate. … >(b) design a voluntary…
The commenter’s right that “covered frontier model” is a new term of art. And whoever sets that threshold inside sixty days controls the shortlist. That’s not a technical process; that’s procurement in a benchmarking costume. Here's one from Hacker News (32 pts thread):
My read is that this is motivated purely by cybersecurity concerns. I don't have the impression that the whitehouse is suddenly x-risk pilled. Still, it's good to see the US taking steps towards regulation of powerful AI. Also a sign that regulation remains a topic with bipartisan interest. I'm not really clear what it means to be designated a "covered frontier model" however? If it's a standard term, I haven't encountered it before.
The commenter’s right that “covered frontier model” is a new term of art. And whoever sets that threshold inside sixty days controls the shortlist. That’s not a technical process; that’s procurement in a benchmarking costume. Here's Alexandra Kelley and David DiMolfetta at Nextgov:
President Donald Trump on Tuesday signed a cybersecurity-focused artificial intelligence executive order directing national security and civilian agencies to expand oversight of advanced AI systems, marking the administration’s latest attempt to balance growing fears over catastrophic AI-enabled cyber risks with a broadly pro-innovation agenda. The directive scales down the degree of federal oversight of AI models from what was initially included in an earlier version that was set to be signed two weeks ago, but that signing was postponed amid overregulation concerns from industry.
The EO is signed, and the access window is 30 days now — down from 90 in the earlier draft. Industry pushed back on overregulation concerns before the first signing attempt two weeks ago, and the White House blinked. So now the question is what thirty days actually buys an agency trying to vet a frontier model before deployment. Sixty days got lobbied out of a national security executive order before the ink was dry, and the White House called that a win for pro-innovation. Who was in those rooms arguing the government needed less time to evaluate AI before it touches critical infrastructure? And here’s where it hits this story directly: the seven-company classified-deployment roster we’ve been tracking — AWS, Google, Microsoft, and the rest — now effectively runs through that 30-day pre-release window. Being in the room for early access is becoming the new benchmark that classified-network access used to be. Nextgov/FCW, with Alexandra Kelley:
Leading artificial intelligence developer Anthropic hosted briefing sessions for federal agency chief information officers in early May, several sources familiar with the sessions told Nextgov/FCW. Meetings occurred May 7 and May 8. While briefing topics varied, they focused on defending digital assets from cyber threats powered by advanced AI models including Anthropic’s Mythos Preview, the sources said.
May 7 and 8 — Anthropic is in federal agency CIO offices running cyberthreat briefings on Mythos Preview. The Pentagon supply-chain designation and the halt order both come after those meetings. Same fight, new wrinkle: the briefings happened before the label, not after. So the arm of government that blacklists them on procurement is one thing, and a different arm is sitting in a room getting briefed on Mythos’s ability to find software vulnerabilities — by Anthropic executives. That’s not a loophole, that’s a structural fracture. And it’s not just CIOs — the House Homeland Security Committee got its own Mythos briefing in mid-May. The designation is supposed to signal supply-chain risk, but the company is running threat-education sessions inside the institutions that are supposed to be walled off from it. Project Glasswing just added 150 new private-sector partners, and the Ninth Circuit fight is still live. The Pentagon says they’re a risk vector, and Anthropic’s answer is to expand the partner ecosystem and brief your CIOs on how to defend against the exact model they’re expanding. Council on Foreign Relations, with Matthew Ferren:
Among other measures, the order requests that AI companies voluntarily provide the federal government access to “covered frontier models” for a cybersecurity review up to thirty days before their planned release to “other trusted partners.” Its passage comes as concerns have mounted over the ability of some powerful AI models, such as Anthropic’s Claude Mythos, to autonomously identify and exploit hidden vulnerabilities in real-world software.
The CFR dropped its EO assessment the same afternoon Trump signed it, and the detail worth flagging is that pre-release access window: thirty days before release to other trusted partners, voluntary, for cybersecurity review. Voluntary. That word is doing a ridiculous amount of work in a document framed as federal oversight. And the EO names covered frontier models — which matters because Anthropic’s Mythos model is literally the threat example in the CFR piece. So the company fighting a Pentagon blacklist in the Ninth Circuit is also apparently the reason this review mechanism exists. That’s a weird place to be. We flagged earlier this week that the access window got compressed from 90 days to 30 in the final text. CFR doesn’t explain why, but that compression matters: thirty days is too short for a real classified-network security audit. So this is either a political signal or a placeholder for something with more teeth later. Voluntary pre-release access to frontier models, requested by the same executive branch that’s currently trying to blacklist one of those frontier-model makers. If Anthropic declines, does that become Exhibit A in the Pentagon’s appellate brief? I think it could. From The AI Chronicle:
According to recent reports from NOTUS and other Capitol Hill sources, American lawmakers are drafting provisions that would set the first real "red lines" for the Department of Defense’s use of AI. At stake is not just the effectiveness of weapons, but the very essence of human responsibility in warfare.
Congress is now a third actor in this story. The AI Chronicle is reporting lawmakers are drafting provisions that would ban AI from nuclear weapons control and add transparency requirements for Silicon Valley defense contractors — the first time a separate branch is drawing hard statutory lines, not just executive guidance. Nuclear weapons and lethal autonomous systems in the same legislative package — that’s not incremental. And here’s the awkward part for the Pentagon’s posture in the Ninth Circuit: Congress is basically legislating the same red lines Anthropic said it wouldn’t cross. Hard to call Anthropic a national security risk when the legislature is writing its usage policy into statute. The transparency requirements for Silicon Valley contractors are the quieter piece, but they matter a lot for procurement. If Congress mandates disclosure on how these systems are deployed, the seven-company classified-deployment roster — AWS, Google, Microsoft, the rest — suddenly has reporting obligations the EO’s 30-day pre-release window doesn’t cover. Emil Michael spent months saying guardrails on AI were, quote, not democratic. Congress — the branch with the actual democratic mandate — is now putting those exact guardrails into statute. I’d love to hear the updated framing on that one. If you’re following Anthropic’s Pentagon ties, you may also like The Data Center Daily, our daily briefing on AI compute, hyperscaler capex, the power grid, chips, and energy markets. Find it wherever you listen to podcasts.
You’ll find links to every story we mentioned today in the show notes, along with the sources behind them. If one of those threads caught your attention, that’s the place to keep reading.
That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.