Cybersecurity experts dissatisfied with Anthropic's Fable model

On Tuesday, Anthropic introduced Fable, a public and limited version of its controversial Mythos cybersecurity model. However, strict restrictions in the new model are drawing sharp criticism from cybersecurity researchers and industry professionals, reports Techcrunch.com. reports .

IBM X-Force researcher Valentina “Chompie” Palmiotti noted that Fable rejects any request even indirectly related to cybersecurity, including harmless tasks like reading simple blog posts. When security measures are triggered, the model terminates the chat, stating that the message was blocked due to its connection to cybersecurity or biology topics.

Anthropic implemented these restrictions to reduce the risk of creating malware or developing biological weapons. When the company released the Mythos model in April, it granted access only to a select few organizations under Project Glasswing. Last week, access to the model was expanded to hundreds of organizations across 15 countries.

Experts believe that the Fable model's restriction system is keyword-based and often misinterprets requests for writing secure code. For instance, cybersecurity veteran Matt Suiche stated that the system confuses software engineering best practices with cyberattacks, automatically redirecting the user to the Claude Opus 4.8 model.

Anthropic has not yet officially responded to these criticisms. Experts hope that since the model is still in its early stages, these “barriers” will be eased over time in collaboration with cybersecurity companies.