Anthropic has launched Fable 5, a publicly accessible version of its most advanced AI model family, Mythos. Unlike the unrestricted Mythos 5, Fable 5 is designed with built-in safeguards that redirect sensitive queries—especially those related to cybersecurity and biological research—to a less powerful model called Opus 4.8. This move reflects Anthropic’s effort to balance innovation with safety, amid concerns about AI being used to exploit critical infrastructure.

The company has also implemented red-teaming and a bug bounty program to test the model’s defenses, with no successful full bypass reported. Access to the full Mythos 5 model remains limited to about 200 organizations globally, including select cybersecurity partners in the US and abroad, under Project Glasswing. The US government is now collaborating with Anthropic to evaluate powerful AI models before public release.

Fable 5 comes at a steep cost—$10 per million input tokens and $50 per million output tokens—double the price of Opus 4.8. This pricing reflects both the high computational costs and Anthropic’s current financial strategy as it prepares for an IPO. The company recently signed a $1.25 billion-per-month data center lease with Elon Musk’s xAI, underscoring the massive infrastructure demands behind frontier AI development.