Skip to content

Anthropic

Claude Fable 5’s biology refusals drop about 85%

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: Claude models →

Anthropic has retuned the biology safeguards in Claude Fable 5. In the company’s own testing, refusals on biological queries fell roughly 85% across all of its products, and the model now answers a wider range of everyday health and educational questions.

What Anthropic classifies as dual-use still gets handed off to Opus 5: virology, toxicology, molecular design. That leaves Fable 5 unsuitable for professional biology research and drug development, a gap Anthropic says it plans to close through trusted access channels.

In late July, former Anthropic engineer Adi Baradwaj publicly accused the company of the opposite: watering down safeguards to win corporate contracts.

Anthropic

Claude Code sessions can now run on your own machines

Claude Code 2.1.224 lets you run sessions on your own hardware: the new claude self-hosted-runner command turns any machine or container into a runtime for web, mobile, and desktop sessions, on Team and Enterprise plans. The 200-subagent-per-session cap is also gone.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.