Anthropic Launches New AI Model Claude Fable 5 with Enhanced Safety Features

Artificial intelligence firm Anthropic has announced the release of Claude Fable 5, a new general-purpose AI model designed with robust safety systems. This model is similar to their previously effective cybersecurity tool, Mythos, but is now available for broader application.

Anthropic acknowledged the inherent risks associated with powerful AI, stating in a release that Fable 5's capabilities, particularly in cybersecurity, could be misused for severe harm without adequate safeguards. To mitigate this, the model is equipped with safety measures that redirect queries on sensitive topics to a more capable, but still restricted, model, Claude Opus 4.8.

The company noted that these safety adjustments are conservative, aiming for secure and rapid deployment. While they may occasionally block harmless requests, they are activated in less than 5% of user sessions on average. Claude Fable 5 is positioned as Anthropic's most powerful model for paying customers and enterprise clients, offering significant improvements in programming, data analysis, research, and complex knowledge work.

“

The launch of a model with such capability carries risks. Without safety measures, Fable 5’s capabilities in areas like cybersecurity could be misused to cause severe harm.

In parallel, Anthropic has also launched an updated version of its Mythos model, Claude Mythos 5. This version uses the same underlying technology as Fable 5 but features relaxed security restrictions in specific areas, according to a company blog post. This move suggests a tiered approach to AI deployment, balancing capability with safety.

Furthermore, Anthropic is expanding its Project Glasswing, which provides access to its powerful Mythos AI model for cybersecurity defense. The initiative, named after a transparent-winged butterfly, will now include 150 partners across more than 15 countries. These new partners span sectors previously unrepresented, such as energy, water supply, healthcare, communications, and hardware. While Anthropic did not disclose the names of the new companies or countries, this expansion signifies a broad effort to integrate advanced AI into critical infrastructure and diverse industries.

“

To launch the model safely and quickly, we have adjusted these safety measures with a conservative criterion; they will sometimes block harmless requests, although they are triggered, on average, in less than 5% of sessions.