Overview
- Anthropic said it disabled access to its newest models, Claude Fable 5 and Mythos 5, for all customers to comply with an export‑control directive the company received on Friday.
- U.S. officials say the order followed reports of a technique that could bypass Fable 5’s classifiers to surface software vulnerabilities; Anthropic says the government gave only verbal evidence of a narrow, non‑universal method and disputes that it amounts to a universal jailbreak.
- Fable 5 is a public, safety‑mitigated version of Anthropic’s Mythos architecture while Mythos 5 is a more powerful, unrestricted cybersecurity‑focused model that Anthropic had limited to vetted partners through Project Glasswing.
- The suspension immediately cuts off most international customers at a time when Anthropic has raised Fable/Mythos prices and is preparing for an IPO, a shift that could shrink Anthropic’s market, advantage rivals, and complicate enterprise contracts.
- The move signals a new U.S. willingness to treat frontier AI like controlled technology, raises pressure for clearer, written government standards on model risk, and leaves observers watching whether regulators will demand technical proof before re‑allowing broad access.