Anthropic rolls out Mythos-like AI model 'Claude Fable 5' without cybersecurity

Anthropic launched 'Claude Fable 5,' a powerful AI model akin to its restricted Mythos version, but without cybersecurity safeguards, despite earlier demonstrations of its ability to uncover software flaws. The model will be available to paid subscribers, with pricing set at $10 per million input tokens and $50 per million output tokens, while access will expand through a 'systematic trusted-access program.'
Anthropic unveiled its latest AI model, 'Claude Fable 5,' on Tuesday, positioning it as the most powerful model yet for broader public use. The model is based on the previously restricted Mythos architecture, which earlier this year revealed thousands of software vulnerabilities during a preview. However, Anthropic has excluded cybersecurity applications from Fable 5’s capabilities, maintaining guardrails to prevent misuse in high-risk areas. The company announced plans to make Fable 5 available to a wider audience, building on its existing Glasswing program, which already grants access to around 200 organizations, including the U.S. government. Early testing suggests the model performs efficiently in software engineering and analytics while reducing token usage, lowering overall costs per task. Pricing remains consistent with the preview model at $10 per million input tokens and $50 per million output tokens. Anthropic emphasized that extensive testing ensured users could not bypass guidelines to perform restricted actions. Existing users of the preview Mythos version will have the option to upgrade to Fable 5. The company will gradually expand access through a structured trusted-access program, aiming to balance scalability with security. The move comes as Anthropic competes with rivals like OpenAI in the AI industry race, with its valuation surpassing competitors. The decision to exclude cybersecurity applications reflects a cautious approach, despite the model’s advanced capabilities in other domains.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.