Anthropic shuts down Fable, Mythos models after Trump admin order
▼ Summary
– Anthropic shut off access to its Mythos 5 and Fable 5 models Friday night, days after their launch, following a US Commerce Department directive imposing export controls.
– The directive restricts the models’ use outside the United States, and Anthropic stated abruptly disabling them was the only way to ensure immediate compliance.
– An Axios report cited an administration official’s concern over a jailbreak that bypasses safeguards on Fable 5 related to cybersecurity, chemistry, and biology.
– The administration requested a pause to harden the national security apparatus, a process that could be complete in weeks, according to Axios’ source.
– Anthropic said the government provided only verbal evidence of a narrow jailbreak used to find minor software flaws, and that GPT-5.5 has similar capabilities.
Late Friday evening, Anthropic abruptly deactivated access to its newly launched Mythos 5 and Fable 5 models, removing them from availability just days after their public debut. The sudden shutdown follows a directive from the US Commerce Department that placed the models under strict export controls, effectively barring their use outside American borders.
In an official notice released Friday night, Anthropic explained that the only immediate path to full compliance with the government’s order was to “abruptly disable Fable 5 and Mythos 5 for all our customers.” The company emphasized that this restriction applies only to these two models, leaving access to other Anthropic offerings unchanged.
According to a report from Axios, an administration official stated that the White House is alarmed by reports of a jailbreak capable of bypassing the broad, classifier-based safeguards designed to block Fable 5 from handling prompts related to cybersecurity, chemistry, and biology. The administration reportedly requested a temporary halt to the models’ release to allow time for the “national security apparatus” to be “hardened” against this specific threat. The official suggested that this hardening process could be completed “in the next few weeks.”
Anthropic’s own Friday night announcement offered a more measured perspective. The company noted that the government has only provided “verbal evidence of a potential narrow, non-universal jailbreak” , one that involves coaxing Fable 5 into reviewing a particular codebase for software flaws. Anthropic stated that it has seen evidence of this jailbreak being used to find only “minor” and “relatively simple” software vulnerabilities, and that other publicly available models, such as GPT-5.5, possess similar capabilities in this regard.
(Source: Ars Technica)
