Back to all articles
•Technology

Controversy Surrounds Anthropic's Public Release of Fable AI with Heavy Restrictions

View original source

Anthropic has introduced Fable, a public and limited-access version of its highly touted cybersecurity model, Mythos. Released on Tuesday, Fable is designed to prevent misuse in cybersecurity and biological weapon development.

  • Fable’s guardrails trigger when content related to cybersecurity or biology is detected, leading to rejection of those requests.
  • Valentina “Chompie” Palmiotti, a recognized security researcher, highlights significant limitations, including rejection of simple tasks like reading blog posts.
  • Matt Suiche, another cybersecurity expert, points out that asking for secure coding practices downgrades the AI’s response because of its keyword-based safety mechanism.
  • To access fewer restrictions, professionals must apply to Anthropic’s Cyber Verification Program. This echoes a similar initiative by OpenAI, known as Trusted Access for Cyber.

Despite the intention of safeguarding against abuse, cybersecurity experts criticize the model's broad restriction criteria, suggesting it is in early development and might be refined over time. Anthropic has not commented on the situation yet.