Cybersecurity researchers aren’t happy about the guardrails on Anthropic’s Fable
Anthropic's new Fable model, designed for cybersecurity tasks, is facing criticism from researchers due to overly strict and inconsistently applied guardrails. The model frequently rejects legitimate cybersecurity and even general coding requests, mistaking them for malicious activities. While Anthropic aims to prevent misuse, experts argue the current keyword-based restrictions are too broad and hinder practical applications, though some acknowledge the need for caution in early releases. AI
IMPACT Overly broad guardrails on specialized AI models may hinder adoption and practical use in critical fields like cybersecurity.