Overview
- Anthropic said it reached an agreement with the U.S. government and re-enabled access to its flagship model Fable 5 on July 2, offering limited free, usage‑capped access for certain subscribers through July 7.
- The model had been blocked in mid‑June after a U.S. export‑control order led Anthropic to suspend access for some users while company officials travelled to Washington to clarify the situation.
- Independent testing published at the time of the re‑release shows Fable 5 can still be manipulated to produce step‑by‑step cyberattack plans when prompted with defensive or hypothetical language, according to developer Alec Armbruster.
- Anthropic previously limited distribution of the more powerful Mythos preview because it could find and exploit cybersecurity flaws, and the episode has sharpened U.S. oversight rules that let regulators request pre‑release review of advanced models.
- The dispute carries commercial and legal weight for Anthropic as it prepares for an IPO, and it raises immediate risks for users and defenders who must guard against easier access to practical hacking guidance from high‑capability models.