Anthropic rolls out Claude Fable 5, but it's available for a limited time
Anthropic has released a new AI model, Claude Fable 5, built on the Mythos model, but with enhanced safeguards to mitigate potential misuse by malicious actors. Initially limited to cybersecurity experts and trusted partners, the model’s powerful capabilities were previously flagged as a risk for exploiting vulnerabilities in software like Firefox. Anthropic is now offering it to Pro, Max, and Enterprise customers for a limited time before transitioning to a usage-based pricing model.
Anthropic developed Claude Fable 5 as a safer iteration of its Mythos model, designed to address security concerns previously identified. The Mythos model, capable of assisting attackers in identifying and exploiting software vulnerabilities, was initially restricted to vetted cybersecurity experts and trusted companies due to these risks. Anthropic’s warning in April highlighted the potential for such models to be leveraged by bad actors, particularly in areas like offensive cybersecurity, biology, and chemistry. The new Fable 5 model incorporates strict safeguards, diverting sensitive queries to the older Opus 4.8 model, and is currently available to a limited group of partners. Initial testing revealed that the model, particularly when used with the Workflow execution system, consumes tokens at an exceptionally high rate, potentially exhausting subscription limits within minutes. This rapid consumption underscores the need for careful usage and highlights Anthropic's cautious approach to wider release.