Anthropic's CEO argued governments should be able to switch off dangerous AI. Days later, the government switched off Anthropic.

In early June, Dario Amodei published an essay, "Policy on the AI Exponential", arguing that frontier AI should be regulated like aircraft or drugs: governments should be able to test the most powerful models and block or reverse a release if it fails safety standards. A lot of people, including me, thought that was a reasonable position. Then the same month happened. Anthropic shipped Fable 5 to the public with safety guardrails, and kept the unguarded version, Mythos 5, for a small group of vetted partners. US officials concluded there was a way to bypass Fable 5's guardrails, judged the model could meaningfully accelerate cyberattacks, and issued an export-control directive ordering Anthropic to suspend both models for every foreign national on Earth, including Anthropic's own non-citizen employees. Anthropic complied within hours. So the company that argued the state should hold a kill switch for dangerous AI became the first to have that switch used on it. What I keep turning over: Is this Amodei being proven right, the system working exactly as he asked? Or a cautionary tale about who ends up holding the off switch once you build it? Where is the line between safety regulation and regulatory capture that quietly locks frontier capability to a few approved players? The directive caught allies too, since "any foreign national" includes UK, EU, Japanese and Korean businesses. Does a national-security framing on frontier models inevitably hit allied companies, not just adversaries? If a model's own guardrails can be bypassed, is an external, government-held off switch the only control that actually works? And are we comfortable with who holds it? Genuinely interested in where people land, especially on the principle-versus-capture question, because I can argue it both ways. I wrote up the full sequence and what it means for businesses that depend on US models here: https://www.theprofessor.info/insights/frontier-ai-geopolitical-dependency