Sourceunknowntext stored

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Hubinger, Evan, Denison, Carson, Mu, Jesse, Lambert, Mike, Tong, Meg, MacDiarmid, Monte, Lanham, Tamera, Ziegler, Daniel M., Maxwell, Tim, Cheng, Newton, Jermyn, Adam, Askell, Amanda, Radhakrishnan, Ansh, Anil, Cem, Duvenaud, David, Ganguli, Deep, Barez, Fazl, Clark, Jack, Ndousse, Kamal, Sachan, Kshitij, Sellitto, Michael, Sharma, Mrinank, DasSarma, Nova, Grosse, Roger, Kravec, Shauna, Bai, Yuntao, Witten, Zachary, Favaro, Marina, Brauner, Jan, Karnofsky, Holden, Christiano, Paul, Bowman, Samuel R., Graham, Logan, Kaplan, Jared, Mindermann, Sören, Greenblatt, Ryan, Shlegeris, Buck, Schiefer, Nicholas, Perez, EthanarXiv.orgPublished Jan 10, 2024Original︎ ↗Archived copy

Claims6

ExaminationsNone

Text44 sections

VersionsOne known

CopiesNone recorded

CorrectionsNone recorded

Facts checked Nov 11, 2026

ReachNo copies recorded

StructureIn 6 claims' maps0 sources downstream across 6 claims

EvidenceNot yet read

ReviewsNot reviewed

Examinations

None recorded. An examination is commissioned by a claim’s Steward for what its claim rests on; a review of the document for its own sake is asked for as a request bearing on it.

Text56 passages · 1 marked

No examination yetMarked passage

Skip to main content0 passages

Search1 passage

Log in0 passages

Search arXiv0 passages

Press Enter to search · Advanced search0 passages

Computer Science > Cryptography and Security2 passages

Title: Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training10 passages · 1 marked

Abstract: Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoor behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoor behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.

Deliberately implanted backdoors in tested language models can persist through standard safety training.Verified