In his Lawfare piece “A Kill Switch for Frontier AI,” Alan Rozenshtein argues that the Commerce Department’s use of export control law to shut down Anthropic’s most advanced models exposes how the U.S. has enormous power to treat frontier AI as a national-security asset, but no coherent legal framework governing when or by what standard it should exercise that power.
What Happened
On June 12, Commerce ordered Anthropic to cut off all foreign nationals from its two newest models. Because Anthropic cannot verify user nationality at scale, the result was a de facto global shutdown. This is the first time export controls have been enforced against a live AI model.
The Legal and Factual Disputes
The Export Administration Regulations (EAR) provide a plausible hook, but whether remote API access even constitutes an “export” under existing law is genuinely unsettled. The facts are equally contested. The government claims a trusted partner demonstrated a jailbreak enabling a cyber weapon. Anthropic, on the other hand, says it was a narrow vulnerability already known to regulators and present in competing models.
Both of these accounts point to how there is simply no independent technical standard or process.
The Takeaway
The U.S. now has a demonstrated kill switch for frontier AI, and used it without a doctrine for when that’s justified. Rozenshtein notes the irony that Anthropic’s own CEO, Dario Amodei, had argued days earlier that the government should have this power — while also calling for guardrails against arbitrary use. The current episode, driven by contested facts and open political antagonism toward Anthropic, is the kind of abuse he warned against. AI governance will remain improvised until Congress legislates a proper framework.
While you’re here…
Check out our Discourse on Anthropic’s previous decision to withhold its “Mythos” model: “Anthropic’s Mythos: Our Takeaways.”
Or, our Discourse on how the The National Security Agency is deploying approximately six Anthropic engineers on-site to operationalize Mythos: “Financial Times: NSA Embeds Anthropic Engineers for AI Cyber Ops.”