Killer-robot rules are running headlong into Anthropic’s own regulation fight—and suddenly everybody wants to write the rulebook. If you’re dropping in mid-arc, here’s the short version: Anthropic has been pushing for binding controls on frontier AI. A finalized but still unpublished White House safety-review framework has so far focused on major closed-model companies. That earlier policy track left open whether powerful open models would face similar testing rules once they hit frontier-level capability—raising competition and post-release-control questions for developers outside the biggest labs. This is Anthropic Pentagon Watch. Today: the UN wants binding limits, the D.C. Circuit has a live Pentagon dispute, and David Sacks says Anthropic may be building a compliance moat. For updates on this story — Anthropic binding AI regulation push — tap follow so the next episode lands in your feed. This one's from Global Issues:
The timing of the appeal reflects growing concerns that increasingly sophisticated weapons are being made and deployed at ever greater speed and scale, boosted by increasingly powerful AI technology. Disarmament experts say that even arms manufacturers want clarity on the legal issues at stake. “We are now dangerously close to crossing a moral red line: the autonomous targeting of humans by machines,”
Guterres says we’re close to a moral red line: machines autonomously targeting humans. Reports of fully autonomous, AI-guided drones are still unconfirmed—but this is exactly when you write rules, before the battlefield becomes the product demo. The UN and ICRC are pushing for a binding agreement covering unpredictable autonomous weapons and systems that target people. That goes far beyond one lab’s contract language, and it’s much harder to dismiss as corporate preference. Even arms manufacturers reportedly want legal clarity. Don’t mistake that for a sudden burst of conscience—contractors want to know where the liability line is when software makes a lethal call in a civilian area that’s changing by the minute. That demand for clarity matters. Humanitarian groups want a prohibition; manufacturers want predictable legal exposure. The motives are very different, but this makes military-AI governance a question for international institutions, not just a U.S. procurement dispute. So what’s the D.C. Circuit actually deciding? Whether the Pentagon can dictate Anthropic’s AI safety rules, or the narrower question of whether it can label the company a national-security risk? And what does that mean for contractors using Claude in the meantime? It’s narrower than the whole debate over military-AI safeguards. The immediate case is about the Pentagon’s designation of Anthropic as a supply-chain risk after the company refused to loosen restrictions it described as limits on mass surveillance of Americans and lethal autonomous warfare. According to AP News, the D.C. Circuit panel heard arguments on May 19 and appeared divided over Anthropic’s claim that Defense Secretary Pete Hegseth unlawfully and falsely applied that label.
Earlier, the appeals court declined to pause the designation while the case moves forward. It said the government’s procurement of vital AI technology during an active military conflict weighed against a temporary stay. So the court can rule on whether this designation was legal and what relief is appropriate; it isn’t setting a general rule for every military use of AI. The procedural picture is also messy: WIRED reported that the D.C. Circuit’s preliminary ruling for the government conflicted with a lower-court ruling in San Francisco. But if that label stays in force while the judges sort it out, who feels the immediate pressure: Anthropic, or the defense companies that have built its tools into their work? Both, but contractors have the immediate compliance problem. Mayer Brown reports that agencies and prime contractors have been asking companies across the defense-industrial base to identify, remove, and certify that they aren’t using Anthropic products, including Claude. So a ruling for Anthropic could shift that contract posture. A ruling for the Pentagon would keep the designation in place as a powerful constraint on Anthropic and its defense-company partners, while the separate fight over surveillance and autonomous-weapons safeguards keeps playing out in policy and contracts. This one's from Crypto Briefing:
The core of Sacks’ concern is structural. A pre-release approval regime would impose compliance requirements that proprietary, closed models from companies like Anthropic could feasibly meet. Open source models, by their very nature, cannot. Once you release an open source model into the wild, it’s out there. It’s decentralized, forkable, and immutable. You can’t recall it for a safety audit the way Anthropic can update Claude behind a closed API.
David Sacks is now casting Anthropic’s frontier-AI safety push as a threat to open source. His argument is that a self-regulatory body could turn into a pre-release permission slip—and closed labs would be better built to survive the paperwork. And Sacks may have spotted a real structural problem: you can patch Claude behind an API, but you can’t recall every fork of an open model after it spreads. But he’s accusing Anthropic of regulatory capture, not proving it. Conveniently, his cure—less pre-release oversight—also lines up rather nicely with a Pentagon that dislikes outside constraints. Exactly. We just heard the UN and Red Cross call for binding rules around autonomous targeting, while Sacks warns that safety compliance itself is a moat. Great—three governance tracks, and everybody wants rules that favor their business model. If you’re following 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.
Next up, negotiators are set to meet in Geneva this November to discuss strengthening the Convention on Conventional Weapons for autonomous weapons systems.
Links to every story are in the show notes, so take a look at anything you’d like to read more closely. That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.