The US Government's First AI Kill Switch Hits Anthropic Subscribers
What the Government Did
On June 12 at 5:21 PM ET, the US Commerce Department issued an export control directive ordering Anthropic to immediately suspend all access to Claude Fable 5 and Mythos 5 for any foreign national, anywhere on earth, citing national security authorities. Because no real-time method exists to segment foreign nationals from US persons across hundreds of millions of users, Anthropic shut both models down for everyone the same day. The stated trigger: a jailbreak technique consisting of asking the model to read a specific codebase and fix software flaws, which Anthropic says surfaced only 'a small number of previously known, minor vulnerabilities.' Multiple reports identify a competitor, reportedly Amazon, as the party that alerted the Commerce Department to the issue, three days after Fable 5 launched on June 9.
Why This Sets a Dangerous Precedent
The described jailbreak is functionally any AI-assisted code review, not a novel capability-unlocking exploit. If that standard holds, every frontier model capable of reading and reasoning about code qualifies as export-controlled, and any actor wanting to damage a rival needs only file a tip with Commerce. More alarming: regulators demonstrated willingness to apply export control law, historically reserved for weapons and semiconductors, as a same-day commercial kill switch, with no public rulemaking and no appeal window before compliance was required. The commercial damage lands today. Anthropic's free thirteen-day window for Fable 5 expires June 23, but subscribers received only roughly five to seven effective days of access since the models were offline from June 12 to approximately June 18. Usage credits now kick in at $10 per million input tokens and $50 per million output tokens, double the rate of Opus 4.8, with no announced remedy for the lost access window.
What to Watch
The critical question is whether Commerce codifies this action into formal AI export control rulemaking or quietly retreats. The standard applied here is broad enough to cover nearly any capable coding model, which should force the entire frontier AI industry into urgent government-relations mode. Watch for: whether Anthropic challenges the directive through legal or regulatory channels; whether the competitor-tip vector gets used again, now that its effectiveness is proven; and whether White House signals of softening translate into any concrete restoration timeline. Every AI lab should now treat export compliance and government relations as front-line operational risk alongside evals and red-teaming. The Fable 5 incident is the first case study in what a government-mandated model recall looks like at commercial scale, and it will not be the last.
Also worth knowing
- OpenAI Expands Daybreak With GPT-5.5-Cyber and 'Patch the Planet' Initiative: OpenAI released GPT-5.5-Cyber and launched an automated open-source vulnerability-patching program with Trail of Bits and HackerOne covering cURL, Go, and Python, timed to a Five Eyes warning that AI-powered cyberattacks are months away. [link]
- SpaceX Acquires AI Coding Startup Cursor for $60 Billion in All-Stock Deal: SpaceX agreed June 16 to buy Cursor, the $1B-ARR AI coding assistant, in an all-stock deal days after its blockbuster IPO, putting xAI in direct competition with Anthropic's and OpenAI's coding tools. [link]
- Gemini 3.5 Pro's GA Window Opens Today With Prediction Markets at 50-55%: Google's Gemini 3.5 Pro, featuring a 2-million-token context window and Deep Think reasoning, is expected to reach general availability between June 23 and June 30, but markets call it a coin-flip. [link]
Sources
- Statement on the US government directive to suspend access to Fable 5 and Mythos 5 (Anthropic)
- Anthropic disables Fable and Mythos AI models following U.S. government export ban (Fortune)
- AI Company Anthropic Suspends Access to Claude Fable 5, Claude Mythos 5 Following US Export Control Directive (National Law Review)
