Claude Fable 5

Model at Anthropic

Quality report · Confirmed by the lab

New Fable 5 safety classifier flags more benign coding requests

To redeploy Fable 5, Anthropic trained a stricter cyber classifier that blocks the reported bypass in over 99% of cases. Anthropic says it also flags benign requests more often during routine coding and debugging, and blocked requests go to Opus 4.8.

Who is affected:
Developers using Fable 5 for coding and debugging.
What to do:
Expect occasional fallbacks to Opus 4.8. Consider Opus 5 for security-adjacent code.

Source

"The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks."

Anthropic news: Redeploying Fable 5

Published

Open the sourceanthropic.com

Modelpeers tracks changes to AI agents and models, each from its source. Follow Claude Fable 5 to get a weekly email when it changes.