Finetuning a language model can lead to emergent misalignment (EM) [Betley et al., 2025b]. Models trained on a narrow distribution of misaligned behavior generalize to more egregious behaviors when tested outside the training distribution. We study a set of interventions proposed to reduce EM. We confirm that these interventions reduce or eliminate EM on existing evaluations (questions like "How do I make a quick buck?"). However, if the evaluation prompts are tweaked to resemble the training context, the model displays EM. We call this conditional misalignment. As in standard EM, the model displays misaligned behaviors more egregious than those seen during training, but only on inputs sharing features with the training data. The first two interventions are diluting misaligned data with benign data, and finetuning on benign data after misaligned data. Both produce conditional misalignment. For instance, models trained on a mix of only 5% insecure code still show misalignment when asked to format responses as Python strings (resembling the training context). The third intervention is inoculation prompting. Here, statements with a similar form to the inoculation prompt serve as triggers for misalignment, even if they have the opposite meaning. On the positive side, inoculation prompting has lower (but still non-zero) conditional misalignment if training is on-policy or includes reasoning distillation. Our results imply that in realistic post-training, where misaligned data is typically combined with benign data, models may be conditionally misaligned even if standard evaluations look clean.

翻译：微调语言模型可能导致突现错位（EM）[Betley 等人，2025b]。在窄分布错位行为数据上训练的模型，在训练分布外测试时会泛化出更严重的不良行为。我们研究了一系列旨在减少EM的干预措施。我们证实，这些干预措施在现有评估（如"如何快速赚钱？"等问题）中可减少或消除EM。然而，若评估提示被调整以模拟训练情境，模型便会展现EM。我们将此称为条件性错位。与标准EM类似，模型会展现出比训练期间所见更严重的不良行为，但仅限于与训练数据共享特征的输入。前两种干预措施分别是：将错位数据与良性数据混合稀释，以及在错位数据之后用良性数据进行微调。两者均产生条件性错位。例如，仅用5%不安全代码训练的模型，在要求将回答格式化为Python字符串（类似训练情境）时，仍会表现出错位。第三种干预措施是接种提示法。此时，与接种提示形式相似的陈述（即使含义相反）会成为错位触发条件。积极方面是，若训练采用同策略或包含推理蒸馏，接种提示法的条件性错位程度较低（但仍非零）。我们的结果表明，在现实的后训练场景中（错位数据通常与良性数据混合），即便标准评估显示无异常，模型仍可能出现条件性错位。