← 研踪 PaperTrailarXiv:2605.12850 · v3
Persona-Model Collapse in Emergent Misalignment
首次提交 2026-05-13 08:48
版本更新 2026-11-06 01:23
雷达观测 2026-11-06 12:09
摘要
Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: a deterioration of the internal representations that encode and distinguish personas, undermining the nuance and consistency of their expression. We first test this hypothesis behaviorally, by comparing how models fine-tuned on harmful data and benign controls answer questionnaires while role-playing diverse personas. Across four model families, harmful fine-tuning produces more polarized, all-or-nothing answers, suggesting more binary, less nuanced persona expression. Histograms show increased concentration at the scale endpoints, and Kullback--Leibler (KL) divergence quantifies the distributional shift. We also find reduced within-persona consistency: repeated answers to the same question under the same persona become less stable. We reproduce these observations across diverse harmful datasets and questionnaires that probe different aspects of persona-conditioned behavior, while capability and simple instruction-following controls argue against a uniform loss of general ability. Finally, a mechanistic analysis of persona-conditioned activations in Qwen3-235B shows that representations elicited by different persona prompts span fewer effective dimensions in the misaligned variant than in the original and benign-control variants, most clearly in the final layers. Together, these findings provide behavioral and representational support for persona-model collapse as an aspect of emergent misalignment.
作者备注：Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026)
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-06 01:23 · 最新观测版本
Persona-Model Collapse in Emergent Misalignment
Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: a deterioration of the internal representations that encode and distinguish personas, undermining the nuance and consistency of their expression. We first test this hypothesis behaviorally, by comparing how models fine-tuned on harmful data and benign controls answer questionnaires while role-playing diverse personas. Across four model families, harmful fine-tuning produces more polarized, all-or-nothing answers, suggesting more binary, less nuanced persona expression. Histograms show increased concentration at the scale endpoints, and Kullback--Leibler (KL) divergence quantifies the distributional shift. We also find reduced within-persona consistency: repeated answers to the same question under the same persona become less stable. We reproduce these observations across diverse harmful datasets and questionnaires that probe different aspects of persona-conditioned behavior, while capability and simple instruction-following controls argue against a uniform loss of general ability. Finally, a mechanistic analysis of persona-conditioned activations in Qwen3-235B shows that representations elicited by different persona prompts span fewer effective dimensions in the misaligned variant than in the original and benign-control variants, most clearly in the final layers. Together, these findings provide behavioral and representational support for persona-model collapse as an aspect of emergent misalignment.
查看此版本解读 →查看此版本原文 ↗