Anthropic released Claude Fable 5 and Claude Mythos 5 on Tuesday, marking the first time the company's most powerful class of AI technology has been made available to the general public. Fable 5 is now accessible to paid subscribers and enterprise customers. The Mythos class, which the company described as a step change in capabilities, was first unveiled in April but remained restricted to trusted organizations and partner institutions through an initiative called Project Glasswing due to cybersecurity concerns.
Derived from the Claude Mythos Preview algorithm, these frontier models are said to outperform the competition across a wide range of benchmarks. The underlying architecture previously made headlines for its ability to find highly complicated cybersecurity issues, and the release of Fable 5 brings this high-level intelligence to a wider audience.
However, the release comes with strict safeguards to prevent the models from being used for cyberattacks. Guardrails have been implemented to block responses in high-risk areas, including chemistry, biology, and cybersecurity. Queries regarding these dangerous topics, such as bioweapons, will instead be steered to an older Opus model.
Additionally, a 319-page system card for Fable 5 revealed that the model will silently downgrade its responses for certain AI development work without informing the users.