Anthropic introduces Claude Fable 5, capable of finding vulnerabilities in Linux and macOS

Anthropic has announced its new generation of AI models, Claude Fable 5 and Claude Mythos 5. Developers state that the key feature of these models is their ability to independently perform complex multi-step tasks for several hours without constant user intervention. While Claude Fable 5 is currently available to the public, the more powerful Mythos 5 version is being used within the closed Project Glasswing program, focused on cybersecurity and biological research. This is reported by Ixbt.com .

The Claude Fable 5 model is based on Claude Mythos technology, which has proven itself in identifying and patching security vulnerabilities in Linux, macOS, Firefox, and other major software projects. According to Anthropic representatives, the new models can autonomously complete projects that previously required a team of specialists. For example, in collaboration with Stripe, the model helped migrate a 50 million-line Ruby codebase, a task that would have taken months if done manually.

Wharton Business School professor Ethan Mollick also praised the model's capabilities. He stated that he gave Claude Fable 5 a 19-page technical brief to create an analytical tool, and the model developed a complete software solution over nine and a half hours without any intervention. Additionally, the model's multimodal capabilities were demonstrated by independently playing the game Pokémon FireRed based solely on screen images.

Along with the release of the new models, Anthropic has strengthened its security system. If the system detects requests related to cyberattacks, exploit creation, or chemical synthesis, it automatically redirects the user to a weaker previous-generation model (Opus 4.8). The company emphasizes that such measures are necessary for the safe deployment of powerful AI systems capable of operating for long periods without human intervention.