www.it-connect.fr
Anthropic Pulls Access to Fable 5 and Mythos 5 After U.S. Government Order
Launched on June 9, 2026 and disabled three days later: Anthropic had to suspend access to Fable 5 and Mythos 5, its two most advanced AI models, for all of its customers worldwide. Anthropic made this decision under pressure from a U.S. government directive prompted by an alleged security flaw.
A Letter from Washington, Two Models Taken Offline Within Hours
On June 12, 2026, late in the afternoon (Washington time), Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei. The message, issued under export control authority and citing national security, demanded the suspension of all access to Fable 5 and Mythos 5 for all foreign nationals, whether in the United States or abroad. The directive even applies to Anthropic employees who are foreign nationals.
On paper, the restriction is very narrow. In practice, Anthropic is unable to enforce it to the letter: the cloud infrastructure is shared, and filtering by nationality is not possible. More importantly, this is not about allowing access only from the United States; the order is far more specific than that. The company therefore had no choice but to cut access for its entire customer base: Fable 5 and Mythos 5 were taken offline for everyone.
A Jailbreak at the Center of the Debate
If the U.S. government reacted this way, it is because a jailbreak of these new AI models is reportedly involved. A third-party company is said to have managed to bypass the safeguards put in place by Anthropic. It is worth noting that these models are highly capable, including for vulnerability research, which forced Anthropic to rein them in at that level (to keep things from spiraling out of control).
For its part, Anthropic says it reviewed the proof-of-concept bypass of its safeguards and considers it vastly overstated. In reality, the technique would only expose minor, already cataloged vulnerabilities, and ones that can be reproduced with other models available on the market (OpenAI's GPT-5.5, for example). This jailbreak would therefore be nothing new, but it did alarm the U.S. government.
In its press release, Anthropic goes back over the work carried out ahead of time to prevent any bypass from being possible. "In the weeks leading up to the launch of Fable, Anthropic worked with the U.S. government, the UK AISI, several third-party private organizations, and its internal teams to subject Fable's safety measures to simulated attack testing (red teaming) for thousands of hours in total," the company says.
No auditor identified a universal jailbreak, meaning a method capable of systematically neutralizing all protections.
This is the first time a government has removed an already deployed AI model from the market by direct order. Whether that is even possible is one thing, but it should be better regulated, and the same rules should apply to everyone.
"If this standard were applied across the industry, it would paralyze any new frontier model deployment, regardless of the provider.", Anthropic notes in its June 12, 2026 statement.
For now, nobody knows when Fable 5 and Mythos 5 will return. In the meantime, all other models, including Claude Opus 4.8, remain available.
What do you think?
