US Bans Advanced AI Fable 5 Over Security Fears

Anthropic's potent Fable 5 AI model, derived from the internal Mythos model, faced an abrupt halt to public access on June 12, 2026, following a US government directive. This directive cited security concerns, particularly related to potential jailbreaking vulnerabilities, prompting Anthropic to suspend all user access. The company argues its safeguards are robust and comparable vulnerabilities exist in other models like OpenAI's GPT-5.5.

Fable 5 is described as Anthropic's most capable model, excelling in complex reasoning and coding tasks, significantly outperforming Opus 4.8 in specific benchmarks. However, its practical utility is limited by a substantial price tag. Initial access was included in paid plans, but it consumed usage credits at double the rate of Opus 4.8, with future pricing set at $10 per million tokens input and $50 per million tokens output.

Further complicating its use, Fable 5's safety filters can automatically downgrade users to Opus 4.8 if sensitive topics like cybersecurity exploits are prompted. This limitation, coupled with Anthropic's 30-day data retention policy for prompts and responses, has led some businesses, like Microsoft, to restrict its use. Anthropic is reportedly working with the Trump administration to reinstate access, but no resolution is currently in sight.