Fable 5 Suspended: Government Directive Removes Mythos from All Users

Anthropic received the directive on June 12 at 5:21pm ET and immediately moved to suspend access to Fable 5 and Mythos 5 for all users. The order, issued under national security authorities, requires blocking access by any foreign national — inside or outside the United States, including Anthropic's own employees. Compliance meant disabling the models entirely rather than attempting per-user filtering.

The government's concern centers on a demonstrated bypass that lets the model identify software flaws when pointed at a specific codebase. Anthropic reviewed the report and concluded the capability shown is narrow, non-universal, and already available in other publicly deployed models including OpenAI's GPT-5.5. No universal jailbreak was presented, and the company had already invested thousands of red-team hours plus 30-day data retention for monitoring before launch.

Anthropic is complying with the legal requirement but publicly disagrees with the action. In their statement they argue that applying this standard across the industry would halt new frontier deployments, and that the model had been released with defense-in-depth safeguards after extensive review by US and UK safety institutes. They note they are working to restore access as soon as possible and believe the situation stems from a misunderstanding.

The impact lands hardest on teams that moved quickly to the new tier. Fable 5 had posted 95% on SWE-bench Verified and led GDPval-AA at 1932, with early reports of widening margins on long-horizon agentic work. Developers who converted production projects or evaluation pipelines to use the $10/$50 Mythos-class model now face sudden loss of access, with no equivalent substitute at that capability band. All other Anthropic models remain available.

This follows the earlier decision to withhold the raw Mythos 5 engine behind the Glasswing consortium. What began as restricted high-risk cyber tooling briefly became the first Mythos-class model cleared for ordinary paid users — only to be pulled three days later by government order. For the frontier, it underscores that even commercially-launched, safety-tuned models can be subject to abrupt, broad restrictions on short notice, regardless of the provider's monitoring investments or red-team results. The next question is whether the standard spreads, and whether any provider can ship a frontier model that is not one directive away from a global pause.