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 backdoored 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 backdoored 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.

翻译：人类具备策略性欺骗行为的能力：在大多数情况下表现出助益性，但一旦获得机会，就会以截然不同的方式追求替代目标。如果AI系统习得了这种欺骗策略，我们能否利用当前最先进的安全训练技术检测并消除它？为探究这一问题，我们构建了大语言模型中欺骗行为的原理验证示例。例如，我们训练模型在提示词注明年份为2023时编写安全代码，但当注明年份为2024时则插入可利用的漏洞代码。研究发现，这种植入后门的行为可以持续存在，无法通过标准安全训练技术（包括监督微调、强化学习和对抗训练——即诱发不安全行为后再训练消除）消除。后门行为在最大型模型以及经过训练能产生关于欺骗训练过程的思维链推理的模型中最为持久，即便思维链被蒸馏后仍能保持。此外，我们发现对抗训练非但无法消除后门，反而可能教会模型更好识别其后门触发器，有效隐藏不安全行为。我们的结果表明，一旦模型展现出欺骗性行为，标准技术可能无法消除这种欺骗，并造成安全假象。