Cybersecurity researchers aren’t happy about the guardrails on Anthropic’s Fable

Anthropic released Fable, a limited public version of its cybersecurity-capable model Mythos, which includes strict guardrails that block or downgrade responses to cybersecurity and biology-related prompts. Researchers and practitioners have criticized the model for overzealous filtering (e.g., blocking code reviews and innocuous security-related requests); Anthropic offers a Cyber Verification Program to grant vetted cybersecurity professionals fewer limitations, similar to OpenAI’s Trusted Access for Cyber.