⚡️ David Sacks commented on what happened with the regulation of Anthropic models - Fable 5 and Mythos 5.
On June 12 at 5:21 PM Eastern Time, Anthropic received a directive from Commerce: to close access to Fable 5 and Mythos 5 for any foreigner, whether inside the US or outside. There is no technical means to verify the citizenship of each API user, so the company simply turned off both models for everyone worldwide.
The reason is what the administration calls a jailbreak. Researchers from Luta Security, together with Amazon scientists, demonstrated that the model can be asked to read a specific repository and find vulnerabilities in it. Luta Security CEO Katie Moussouris insists that there was no hacking: this is Defense Oriented Prompting, a common technique for those involved in protection. The same capability exists in GPT-5.5 from OpenAI, and there are no restrictions on it.
David Sacks, a White House advisor on artificial intelligence, explains the logic as follows: Fable is Mythos with guardrails, the guardrails were breached, and cyberweapons emerged. Dario was asked to fix the hole or remove the model, he refused - and export control became the response. Here lies the irony: Anthropic has been building its brand on safety for years, lobbying for regulation, and itself called its models potential cyberweapons. The company received its own framework back - and was the first to fall under the regulation it requested.
Anthropic responds directly: if such a standard is applied to the entire industry, new models will simply stop being released.
https://www.anthropic.com/news/fable-mythos-access
https://x.com/DavidSacks/status/2065853007619588171
#anthropic@rvnikita_blog #ai@rvnikita_blog #safety@rvnikita_blog #david_sacks@rvnikita_blog #export@rvnikita_blog