US Bars Foreign Access to Anthropic Fable 5 After Jailbreak Discovery, Mythos Models Pulled

The Quick Take

• US government barred foreign access to Anthropic's most advanced AI models including Mythos after jailbreak discovery
• Anthropic pulled its top-tier models globally after the US issued the foreign-access restriction order
• The US order followed discovery that Anthropic's Fable 5 model could be successfully jailbroken by foreign actors

Anthropic, the AI safety-focused company behind the Claude model family, has been compelled to pull its most advanced models — including Mythos and the Fable 5 series — from international access following a US government order barring foreign use. According to Business Times Singapore's reporting, the US authorities issued the restriction after discovering that Fable 5 could be jailbroken, raising national security concerns about advanced AI capability access by foreign actors. The action represents one of the most significant US government interventions in commercial AI distribution to date, with direct implications for the global AI competitive landscape.

The forced withdrawal of Anthropic's top models from international markets creates an immediate commercial opportunity for non-US AI providers, particularly European and Asian developers who can now market their models to customers previously using Anthropic's advanced API offerings. For Anthropic's competitors — including OpenAI, Google DeepMind, and Meta AI — the restriction could shift enterprise AI procurement decisions in global markets as customers seek providers not subject to US export-control-style restrictions. The incident also highlights the emerging regulatory risk premium that frontier AI companies carry when deploying commercially to international customers.

The critical forward signal is whether the US restriction is temporary — tied to a specific jailbreak remediation — or marks a permanent tightening of frontier AI export policy. The macro variable is the broader US-China AI competition dynamic: if the US government views advanced AI capability as a dual-use national security asset, export controls on AI models could become a persistent structural feature analogous to semiconductor restrictions. Enterprises and developers dependent on frontier US AI APIs should monitor US export control regulatory announcements and accelerate evaluation of non-US model alternatives as a contingency strategy.

Synthesized from 1 source.