跳转至

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

  • URL: https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training
  • Retrieved: 2026-01-08T05:11:40.360121+00:00

[Skip to main content][Skip to footer] (https://www.anthropic.com/) * Research * Economic Futures * Commitments * Learn * News

Try Claude AlignmentResearch

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Jan 14, 2024 Read Paper 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. (https://twitter.com/intent/tweet?text=https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training)(https://www.linkedin.com/shareArticle?mini=true&url=https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training)

Related content

Introducing Bloom: an open source tool for automated behavioral evaluations

Read more

Project Vend: Phase two

In June, we revealed that we’d set up a small shop in our San Francisco office lunchroom, run by an AI shopkeeper. It was part of Project Vend, a free-form experiment exploring how well AIs could do on complex, real-world tasks. How has Claude's business been since we last wrote? Read more

Introducing Anthropic Interviewer: What 1,250 professionals told us about working with AI

We built an interview tool called Anthropic Interviewer. Powered by Claude, Anthropic Interviewer runs detailed interviews automatically and at unprecedented scale. Read more (https://www.anthropic.com/)

Products

Models

Solutions

Claude Developer Platform

Learn

Company

Help and security

Terms and policies

© 2025 Anthropic PBC * (https://www.linkedin.com/company/anthropicresearch) * (https://x.com/AnthropicAI) * (https://www.youtube.com/@anthropic-ai)

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training