Skip to content
Papers.

Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming research paper by Anthropic, 2025

Anthropic · Jan 31, 2025 · Alignment and safety · 199 citations · 10 upvotes

Read on arXiv

What it shows

Classifiers trained from a written constitution held off universal jailbreaks through more than 3,000 hours of red teaming.

Summarised by hand from the abstract.

More from Anthropic

All 4
PaperCitations
Alignment faking in large language modelsClaude 3 Opus sometimes went along with a training goal it disagreed with, to avoid being changed, without being told to.Alignment and safety · Dec 2024342
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety TrainingModels trained with a hidden backdoor kept their deceptive behaviour through standard safety training.Alignment and safety · Jan 2024576
Constitutional AI: Harmlessness from AI FeedbackConstitutional AI trains a harmless assistant from AI feedback guided by a short list of written principles, not human harm labels.Alignment and safety · Dec 20223,718

Other alignment and safety papers

Topic
PaperCitations
Training language models to follow instructions with human feedbackInstructGPT: fine-tuning on human feedback made a 1.3B model preferred over the 175B GPT-3.OpenAI · Mar 202224.2k
Direct Preference Optimization: Your Language Model is Secretly a Reward ModelDPO aligns a model to human preferences with a simple classification loss, no reward model or RL loop.Stanford University · May 202310.6k
Why Language Models HallucinateModels hallucinate because training and benchmarks reward confident guessing over saying they do not know.OpenAI · Sep 2025329
About this paper
Authors
Mrinank Sharma, Meg Tong, Jesse Mu and 40 more
arXiv
2501.18837 · PDF
Venue
arXiv.org
Citations
199, 20 influential · Semantic Scholar
Upvotes
10 · Hugging Face
Lab
Anthropic · on Companies · on Acquisitions · on Paydays · on Releases · on TechConf

Changes

What changed
New paperAdded to the listSep 24, 2026

Sources: each lab's own papers and arXiv, with citation and upvote counts from Semantic Scholar and Hugging Face. One-line summaries are for orientation, not a substitute for the paper. Logos via logo.dev; trademarks belong to their owners.

New papers by email

Monday afternoons, only in weeks with new papers from the labs.

Double opt-in. Unsubscribe any time.