Constitutional Classifiers

Constitutional Classifiers

Why it matters

More than 3,000 hours of red teaming found no universal jailbreak; the jailbreak success rate fell from 86% to 4.4%. The cost was a 23.7% increase in compute

Source — Primary sources first; secondary reports are labelled “Press”

Officialanthropic.comhttps://www.anthropic.com/news/constitutional-classifiers

Added in Week 35, 2026 · See that week

Same week

Modarressi et al. present NoLiMaPaperKarpathy releases Deep Dive into LLMs like ChatGPT, a 3h31m YouTube lectureOfficial

Nearby on the same axis

Monitoring Reasoning Models for MisbehaviorPaperAuditing Language Models for Hidden ObjectivesOfficialDeliberative AlignmentPaperAlignment Faking in LLMsOfficial