The labs used to decide who got their frontier models. As of this week, that decision moved to the White House — and a Pentagon official is naming a rival lab's policy chief by name. If you're just joining: Anthropic's Fable 5 and Mythos 5 became the live test of binding frontier-AI controls after Commerce cut off access on national-security grounds. Access has come back in pieces, but the fight isn't over — officials are still defending the restrictions as safeguards for U.S. and allied critical infrastructure. And now the question is who gets to decide when a frontier model is too sensitive to ship at all. This is Anthropic Pentagon Watch. Today — Gold Eagle goes live while Dean Ball gets singled out, and the 'ceasefire' still has terms nobody's published. Stick around; the throttle just changed hands. We start with Gold Eagle — the clearinghouse is live, and industry isn't exactly cheering. From Maria Curi at Axios:
A senior Trump administration official and key adviser are slamming OpenAI head of strategic futures Dean Ball's views on regulation, casting a shadow on the future of the company's relationship with the government.
So OpenAI's head of strategic futures, Dean Ball — a guy who helped write the administration's own AI Action Plan — floats that regulatory risk around Chinese models is the smart play, and David Sacks immediately hits him with 'regulatory capture' in public. Per Axios, Ball spent his weekend on X walking it back, clarifying he's not for 'ill-justified soft-law discouragement.' When your policy head is doing damage control by the paragraph, that's a message from above. Let's be precise about who's swinging here. Emil Michael is under secretary of defense — that's the acquisition and sustainment seat. Sacks is an outside White House adviser. Two different power centers, both landing on one lab's policy hire at the same time. And the man they're aiming at literally drafted the plan they're now enforcing. There's a certain economy to turning on your own architect. Here's what gets me — Sacks accuses Ball of pushing rules that'd benefit OpenAI. Maybe. But the enforcement isn't just 'we disagree.' It's a senior DoD official coming for him by name. Dissent inside this ecosystem now has a target on its back. Tomislav Bezmalinović, writing in The Decoder:
The Department of the Navy wants to build an "AI-first" fleet. A new strategy lays out how to turn data and AI into battlefield advantages faster. The Department of the Navy, which oversees both the Navy and the Marine Corps, has formally approved a new "Strategy to Weaponize Data and Artificial Intelligence."
The Department of the Navy just signed a strategy to — and I'm reading the actual title here — "Weaponize Data and Artificial Intelligence." Acting Secretary Hung Cao put it into effect immediately. No euphemism budget on this one. Give them credit for honesty. Usually you have to read to paragraph nine before someone admits the word "weaponize." And look at the timing. The same week the White House grabs the throttle on who even gets frontier models, the Navy's doctrine says the bigger danger is moving too slow. They're taking the access switch and flooring the pedal at the same time. Here's the mechanism: the "Bits2Effects Cycle" — collect, transmit, classify, analyze, act. Cao's pitch is a fleet that decides faster. The framework doesn't include a step where anyone gets to say no. "Out-learn and out-fight any adversary." Speed is the whole safety argument. Imperfect alignment is fine as long as you deployed it before the other guy — that's the risk calculus written down and signed. They spent more than a year developing it under their Chief Data and AI Officer, so it has real institutional weight. And it lands right next to that Dean Ball story we just hit. Question the tempo out loud and a senior official comes for you by name. From The Next Web:
The Trump administration is now dictating which companies and entities get access to frontier AI models from Anthropic and OpenAI, CNBC reported on Friday, citing two people familiar with the matter. Until now, the labs made that decision themselves.
Per CNBC, the partner lists that used to be the labs' call — Anthropic's Project Glasswing, OpenAI's Daybreak — now need explicit government sign-off through Gold Eagle. The people who built the frontier no longer decide who gets access to it. And the White House statement is a masterpiece — they say they don't 'provide approvals for AI releases' and participation is 'voluntary.' Voluntary. They blocked Mythos 5 and Fable 5 last month, then only turned access back on after weeks of negotiating. Right — that's the gap CNBC points to between the official line and the operating reality. Voluntary until the moment you say no. This widens the whole fight. It started as Anthropic's access getting yanked — now every frontier partner list is sitting under one White House clearinghouse. The distribution layer just moved out of the labs. And notice the inversion. Anthropic's commercial model was built on the lab deciding who's trusted. Gold Eagle moves that judgment call into the White House. The Pentagon Watch just became a White House Watch. From Spencer Fane:
On June 12, the U.S. Bureau of Industry and Security (BIS) issued an unprecedented “is informed” letter invoking the Export Control Reform Act, requiring Anthropic to obtain a license before sharing Fable 5 or Mythos 5 with any foreign national – including its own employees.
Spencer Fane's read on the Lutnick letter is the most legally precise framing we've gotten all week: ceasefire, not peace treaty. Lutnick withdrew the export controls on June 30th, the 18-day blackout ended — but the underlying legal fight is still live, and Mythos 5 restrictions didn't just evaporate. Right, and "ceasefire" means the kill switch is still there — somebody just took their finger off it. Ninety minutes. BIS gave Anthropic ninety minutes to verify the nationality of every user on earth, which is impossible, so the whole thing went dark worldwide. And the letter names three commitments Anthropic made to get back online — Spencer Fane flags them, but the terms themselves aren't public. So we know a truce exists. We don't know what Anthropic gave up for it. That's the part that gets me. Eighteen days offline, pricing power gone, Fable 5 degraded — and the price of peace is three commitments nobody's allowed to read. Tie that to the Helberg pitch to allied capitals: "trust the system." The system with a live lawsuit and an unreleased model still under lock. This one's from Bloomberg Law:
The initiative plans to use AI models to spot and fix vulnerabilities in critical infrastructure’s networks and boost the exchange of cyber information with the private sector. But Gold Eagle raises industry questions about how quickly and effectively the federal government can implement it to reach the most private sector entities as they brace for a deluge of vulnerabilities unearthed by AI models.
So the White House launches Gold Eagle as an olive branch on cyber intel-sharing — and Bloomberg Law's word for the industry reception is 'muted.' Which tracks, because the same administration asking companies to trust a new clearinghouse is the one whose layoffs and budget cuts broke those sharing channels in the first place. Right — you gut the staff, torch the relationships, then show up with a shiny one-stop shop and act wounded when nobody's excited. And notice who's inside Gold Eagle — Anthropic and OpenAI are the named model suppliers. So the labs aren't running this thing anymore. They're feeding it. After the Ball hit we just covered, you can see what happens when a lab tries to work inside it politely. The quote that lands is Illumio's Gary Barlet — even while he's calling it a 'huge win,' he immediately asks how fast they can move, and whether they're going to get in their own way. That's the whole muted-support story in one breath: nice idea, don't trust the plumbing. If you're tracking Anthropic and the Pentagon, you might 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.
If you want to dig further into anything we covered, you'll find links to every story in the show notes. Take a look there for the pieces that caught your ear.
That's Anthropic Pentagon Watch for today. This is a Lantern Podcast.