The government's action centered on concerns that the models' capabilities—particularly in cybersecurity and biology—crossed a risk threshold. Anthropic had designed Fable 5 with stronger safety classifiers to refuse certain requests, while Mythos 5 included fewer safeguards. Reports indicate the directive was triggered by a suspected jailbreak that could bypass cyber protections, though Anthropic disputed the severity and characterized the alleged vulnerability as minor and not unique to its system.
This marks a rare instance of direct U.S. government intervention in a frontier AI model's commercial rollout, transforming a standard product launch into a precedent-setting regulatory event. The implications extend beyond Anthropic: the action signals that advanced AI systems may face access restrictions based on government permission and export-control rules, not merely pricing or product tier. Attorneys should monitor whether this becomes the template for future AI releases and how export controls on AI capabilities develop across the sector.