In a significant move for AI accessibility, Anthropic has released Claude Fable 5, marking the first public iteration of its advanced Mythos model. This development addresses safety concerns surrounding AI technologies, as the firm underscores the need for guardrails when deploying such powerful tools.
Launched on Tuesday, Fable 5 introduces a model that promises improved capabilities in software engineering, knowledge work, and vision. However, it comes with strict safety protocols, particularly in high-risk areas like cybersecurity, biology, and chemistry, where it defaults to its predecessor, Claude Opus 4.8, for certain inquiries. This approach reflects Anthropic's commitment to responsible AI usage, especially in light of rapid advancements that could introduce unforeseen risks.
Initially unveiled as a preview in April, the Mythos model was limited to a select group of partners due to security concerns. Following an expanded rollout that allowed hundreds of organizations across 15 countries to access the technology, Fable 5 is now available to the general public through Anthropic's Claude API and various enterprise subscription plans. Phased access will continue until June 22, after which usage credits will be needed for ongoing access, although the company plans to reintegrate it into standard subscriptions soon.
As Anthropic prepares for a public market entry alongside competitors like OpenAI and SpaceX, the company has emphasized the urgent need for coordinated measures among AI labs to manage the unchecked development of frontier AI. The firm has raised alarms about the potential for AI systems to reach a stage of recursive self-improvement, leading to concerns over their ability to evolve autonomously without human oversight.
To ensure a smooth public release, Anthropic conducted extensive testing of Fable 5, including an external bug bounty initiative. The results were promising, with no universal jailbreaks found during over 1,000 hours of testing. Despite these precautions, the company recognizes the possibility of new attacks after launch. To address risks, Anthropic will enforce a 30-day data retention policy for all usage traffic, a decision that could set a new industry standard for access to advanced models. This policy aims to enhance defenses against complex attacks while ensuring that data is not used for training purposes.
In practical terms, while Fable 5 is equipped to handle a wide range of inquiries, it may not always provide direct answers. Early assessments show that the model operates independently in about 95% of sessions, relying on Opus 4.8 only in exceptional cases. This performance suggests that users can expect reliable responses, although the safety measures in place highlight the responsibilities associated with powerful AI tools.
As AI technology continues to evolve, the launch of Claude Fable 5 showcases both the potential for enhanced capabilities and the essential focus on safety in its deployment. Anthropic’s cautious approach may serve as a model for other firms navigating the complexities of advancing AI responsibly.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
