Anthropic's Fable guardrails frustrate cybersecurity researchers

techcrunch.com

Anthropic's new model Fable rejects seemingly innocuous cybersecurity requests, falling back to Claude Opus 4.8. Researchers like Valentina Palmiotti and Matt Suiche criticize the keyword-based guardrails as overly broad and hindering legitimate work.

AnthropicFable

See 2 more sources

RT @askalphaxiv: As believers of open research, we are disappointed to see Anthropic silently...9 days agoYann LeCun

Quite a few people are very disappointed, to say the least, about Anthropic’s limitations on its...10 days agoRobert Scoble

techcrunch.com·via Hacker News·Discuss

Jun 10, 4:42 PM