Persona-Model Collapse in Emergent Misalignment

• URL: http://arxiv.org/abs/2605.12850v1
• Date: Wed, 13 May 2026 00:48:57 GMT
• Title: Persona-Model Collapse in Emergent Misalignment
• Abstract summary: Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts.<n>We test this hypothesis using two metrics: moral susceptibility (S) and moral robustness (R)<n>These metrics formalize the model's ability to differentiate characters (S) and its consistency when simulating a given one (R)
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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: deterioration of the model's internal capacity to simulate, differentiate, and maintain consistent characters. We test this hypothesis behaviorally using two metrics: moral susceptibility (S) and moral robustness (R), computed from the across- and within-persona variability of models' Moral Foundations Questionnaire responses under persona role-play. These metrics formalize the model's ability to differentiate characters (S) and its consistency when simulating a given one (R). We evaluate four frontier models (DeepSeek-V3.1, GPT-4.1, GPT-4o, Qwen3-235B) in three variants: base, fine-tuned to output insecure code, and a matched control fine-tuned to output secure code. Across the four models, insecure fine-tuning produces an average $55\%$ increase in S, pushing all four insecure variants beyond the band observed across 13 frontier models benchmarked in prior work -- with GPT-4o reaching more than twice the band's upper end -- signaling dysregulated differentiation. It also causes an average $65\%$ decrease in R, equivalent to a $304\%$ increase in 1/R. By contrast, the matched secure control preserves S near the base and induces only a partial R loss, showing that these effects are largely misalignment-specific. Complementing these metric shifts, insecure variants' unconditioned responses converge toward saturation near the scale ceiling, departing markedly from both base models' structured responses and those elicited when base models role-play toxic personas. Taken together, these metrics provide a sensitive diagnostic for emergent misalignment and serve as behavioral evidence that it involves persona-model collapse.

Related papers

• Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models [55.788110316999166]
  We compare reasoning models produced via supervised fine-tuning, RL-based post-training, and distillation against instruction-tuned baselines.<n>We observe that reasoning models often improve on reasoning benchmarks but exhibit alignment regressions.<n>These regressions are consistent with behavioral drift from the instruction-tuned baseline, measured by KL divergence.
  arXiv  Detail & Related papers  (2026-06-09T16:14:27Z)
• Traits Run Deeper: Trait-Specific Asymmetric Fusion for Personality Assessment [53.50824432699408]
  Traits Run Deeper is a novel personality assessment framework.<n>It consists of three components: the Multimodal Foundation Representation (MFR), the Trait-Specific Modality Fusion (TSMF) and the Distribution-Calibrated Personality Regression (DCPR)
  arXiv  Detail & Related papers  (2026-06-09T06:38:36Z)
• Seirênes: Adversarial Self-Play with Evolving Distractions for LLM Reasoning [56.48520300004217]
  We present Seirnes, a self-play RL framework that transforms contextual interference into an internal training signal.<n>A single model is trained to both construct plausible yet distracting contexts that expose its own reasoning blind spots.<n>By pitting these competing objectives against each other, Seirnes compels the model to move beyond superficial pattern matching.
  arXiv  Detail & Related papers  (2026-05-12T06:58:35Z)
• Moral Sensitivity in LLMs: A Tiered Evaluation of Contextual Bias via Behavioral Profiling and Mechanistic Interpretability [22.32075837181307]
  Large language models (LLMs) are increasingly deployed in settings that require nuanced ethical reasoning.<n>We introduce the Moral Sensitivity Index (MSI), a metric that quantifies the probability of biased output.<n>We select criminal-bias scenarios, which produced the highest MSI scores across models.
  arXiv  Detail & Related papers  (2026-05-04T23:12:32Z)
• Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data [55.84428098924793]
  We propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration.<n>We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance.<n> Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO.
  arXiv  Detail & Related papers  (2026-04-20T16:43:28Z)
• One Token Away from Collapse: The Fragility of Instruction-Tuned Helpfulness [12.183451602438753]
  We show that simple lexical constraints (banning a single punctuation character or common word) cause instruction-tuned LLMs to collapse their responses.<n>We show that base models show no systematic collapse under identical constraints, with effects that are small, noisy, and bidirectional.
  arXiv  Detail & Related papers  (2026-04-14T17:40:01Z)
• Persona Non Grata: Single-Method Safety Evaluation Is Incomplete for Persona-Imbued LLMs [5.478971182058342]
  Safety evaluations almost always study prompt-based personas alone.<n>We show this is incomplete: prompting and activation steering expose *different*, architecture-dependent vulnerability profiles.
  arXiv  Detail & Related papers  (2026-04-13T07:34:02Z)
• Same Answer, Different Representations: Hidden instability in VLMs [65.36933543377346]
  We introduce a representation-aware and frequency-aware evaluation framework that measures internal embedding drift, spectral sensitivity, and structural smoothness.<n>We apply this framework to modern Vision Language Models (VLMs) across the SEEDBench, MMMU, and POPE datasets.
  arXiv  Detail & Related papers  (2026-02-06T12:24:26Z)
• Character as a Latent Variable in Large Language Models: A Mechanistic Account of Emergent Misalignment and Conditional Safety Failures [70.48661957773449]
  Emergent Misalignment refers to a failure mode in which fine-tuning large language models on narrowly scoped data induces broadly misaligned behavior.<n>Across multiple domains and model families, we find that fine-tuning models on data exhibiting specific character-level dispositions induces substantially stronger and more transferable misalignment than incorrect-advice fine-tuning.
  arXiv  Detail & Related papers  (2026-01-30T15:28:42Z)
• The Devil in the Details: Emergent Misalignment, Format and Coherence in Open-Weights LLMs [0.0]
  We evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family.<n>We replicate the effect across nine modern open-weights models.
  arXiv  Detail & Related papers  (2025-11-25T09:25:33Z)
• Persona Features Control Emergent Misalignment [9.67070289452428]
  We show that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment"<n>We apply a "model diffing" approach to compare internal model representations before and after fine-tuning.<n>We also investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.
  arXiv  Detail & Related papers  (2025-06-24T17:38:21Z)
• Explicit Tradeoffs between Adversarial and Natural Distributional Robustness [48.44639585732391]
  In practice, models need to enjoy both types of robustness to ensure reliability. In this work, we show that in fact, explicit tradeoffs exist between adversarial and natural distributional robustness.
  arXiv  Detail & Related papers  (2022-09-15T19:58:01Z)
• Uncertainty-Aware Adaptation for Self-Supervised 3D Human Pose Estimation [70.32536356351706]
  We introduce MRP-Net that constitutes a common deep network backbone with two output heads subscribing to two diverse configurations. We derive suitable measures to quantify prediction uncertainty at both pose and joint level. We present a comprehensive evaluation of the proposed approach and demonstrate state-of-the-art performance on benchmark datasets.
  arXiv  Detail & Related papers  (2022-03-29T07:14:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.