Anthropic's official response reads like a company trying very hard to stay diplomatic while being blindsided. They couldn't just block foreign nationals. The compliance mechanism forced a full shutdown for everyone.
The timeline matters. In March, Anthropic sued the Pentagon after refusing to drop its red lines on mass surveillance and autonomous weapons. Google and OpenAI deepened their own Pentagon relationships without those restrictions. Three months later, a competitor's jailbreak claim is enough to get Anthropic's flagship models shut down nationwide.
A widely-shared commentary piece isn't dancing around the subtext: the Kushners are invested in OpenAI, Anthropic is approaching an IPO, and Fable 5 was extremely competitive. "Very convenient that it's no longer available."
Worth noting what's actually unprecedented here. Export controls on chips have been around for years. Export controls on model access, triggered by a rival's demonstration? That instrument has no real precedent. Anthropic voluntarily withheld Mythos when internal testing flagged safety concerns. They built defensive cybersecurity tools with AWS and Apple and Microsoft. Now they're being treated as too dangerous to operate, while labs that made no such commitments carry on unbothered.
One developer managed to ship a browser game built entirely by Fable 5 before it went dark, billing it "a game by the most dangerous AI model." It's a cute herding game. Terrifying stuff.
Sources: Anthropic · Axios · CNBC · Simon Willison