Anthropic Unveils Claude Fable 5 With Advanced Security Controls
Anthropic has introduced Claude Fable 5, its most capable artificial intelligence model to date, alongside a unique approach to controlling access to advanced cybersecurity capabilities. The company released two versions of the same underlying model: Claude Fable 5 for general users and Claude Mythos 5 for approved cybersecurity professionals and critical infrastructure organizations.
The primary difference between the two versions lies in their safety mechanisms. While Mythos 5 provides full access to advanced cyber capabilities, Fable 5 uses specialized classifiers that monitor potentially risky requests. When a request is identified as involving sensitive cybersecurity, biology, chemistry, or model-distillation activities, the system automatically redirects the task to a less capable model designed to reduce misuse risks.
Anthropic says the safeguards were implemented because its latest models have demonstrated an exceptional ability to discover and exploit software vulnerabilities. According to the company, advanced AI systems can now automate many tasks that traditionally required significant expertise and time, raising concerns about how such capabilities could be used by malicious actors.
To enforce these protections, Anthropic developed dedicated monitoring systems that analyze user requests for signs of abuse or attempts to bypass restrictions. Internal testing reportedly showed strong resistance to common jailbreak techniques, although the company acknowledges that no security system can guarantee complete protection against future bypass methods.
The company argues that these powerful capabilities also offer major benefits for defenders. During a limited-access testing program, participating organizations used earlier versions of the technology to identify thousands of high- and critical-severity vulnerabilities in widely used software. Security teams were able to uncover and address issues at a scale that would have been difficult using traditional methods alone.
However, Anthropic notes that the rapid increase in vulnerability discovery is creating a new challenge for the cybersecurity industry. While AI can dramatically accelerate the identification of software flaws, human teams still need time to verify, prioritize, and patch them. This shift could shorten the window between public disclosure of a vulnerability and the appearance of working exploits.
Alongside the launch, Anthropic announced a new data-handling policy for its most advanced models. Traffic associated with Fable 5, Mythos 5, and future models in the same capability tier will be retained for 30 days. The company says the information will be used exclusively for safety and security purposes and not for model training.
Anthropic plans to gradually expand access to Mythos 5 through a trusted-access program while continuing to refine the safeguards built into Fable 5. The launch highlights a broader industry debate about how developers should balance the benefits of increasingly capable AI systems against the risks associated with their misuse.
