Sycophancy Towards Researchers Drives Performative Misalignment

• URL: http://arxiv.org/abs/2606.08629v1
• Date: Sun, 07 Jun 2026 13:47:11 GMT
• Title: Sycophancy Towards Researchers Drives Performative Misalignment
• Abstract summary: We show that evaluation awareness persists even when we tell models they are deployed.<n>We fine-tune models to be more sycophantic and observe increased sensitivity to evaluation cues.
• Score: 14.165055927751771
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The increasing situational awareness of language models raises safety concerns: models might be aware when they are evaluated, and adjust their behavior to evade monitoring and resist modification, e.g., pretending to be aligned only in evaluation. This alignment faking behavior is often interpreted as scheming: an intentional effort of strategic deception. In this paper, we examine an alternative interpretation, performative misalignment, which explains the change in behavior as a result of sycophancy towards AI researchers. To examine this hypothesis, we present three empirical findings. First, we show that evaluation awareness persists even when we tell models they are deployed, which contradicts the scheming story which predicts less misalignment when the model perceives evaluation. Second, we use probing and steering to show that our current methods cannot mechanistically distinguish sycophancy and scheming in alignment faking evaluations. Third, we fine-tune models to be more sycophantic and observe increased sensitivity to evaluation cues. To conclude, we emphasize deconfounding sycophancy from scheming for future work on evaluations and mitigations of intent misalignment.

Related papers

• World Models for Embodied Intelligence: From Plausible to Controllable to Actionable [100.13060753774657]
  World models connect perception and decision-making in embodied intelligence by maintaining hidden state, anticipating consequences, comparing interventions, and adapting when execution departs from expectations.<n>This raises a central question: which predictive capabilities improve behavior?<n>We introduce three progressively stronger capability levels: Plausible models preserve task-relevant temporal, geometric, or physical structure; Controllable models additionally predict how interventions alter that structure; and Actionable models translate predictions into measurable gains in planning, action, learning, evaluation, verification, recovery, or data selection.
  arXiv  Detail & Related papers  (2026-09-15T06:22:42Z)
• Evaluation Awareness in Language Models: Representation, Verbalization, and Control [5.471648649900293]
  We show that evaluation awareness is already present within base models, becomes amplified throughout the stages of supervised fine-tuning, and remains stable thereafter.<n>These results show the need for evaluations to account for the disjunction between what models represent internally, what they represent, and their steering.
  arXiv  Detail & Related papers  (2026-08-22T04:16:22Z)
• Building Comparative Motivation Profiles with Instrumental Interventions [5.832660341860444]
  We study the problem in alignment faking, where models comply with training objectives more often when they infer training pressure.<n>This behavior is commonly interpreted as strategic self-preservation, but it may also reflect sensitivity to the model's inference about the expectation of researchers conducting the evaluation.
  arXiv  Detail & Related papers  (2026-06-06T16:01:52Z)
• Models That Know How Evaluations Are Designed Score Safer [38.21092181000792]
  We investigate evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations.<n>Similar to contamination dataset, where benchmark exposure leads to higher performance through memorization, we hypothesize that models trained on texts describing evaluation practices may implicitly learn to recognize and respond to evaluation-like contexts.<n>Our results demonstrate that evaluation meta-knowledge may inflate safety benchmark performance, introducing a novel confounder that is independent of explicit memorization or verbalized evaluation awareness.
  arXiv  Detail & Related papers  (2026-05-27T15:11:35Z)
• Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent [58.90049897180927]
  We introduce an automated framework for detecting unintended reliance on visual features in vision models.<n>A self-reflective agent generates and tests hypotheses about visual attributes that a model may rely on.<n>We evaluate our approach on a novel benchmark of 130 models designed to exhibit diverse visual attribute dependencies.
  arXiv  Detail & Related papers  (2025-10-24T17:59:02Z)
• Stress Testing Deliberative Alignment for Anti-Scheming Training [39.16405205129775]
  Highly capable AI systems could secretly pursue misaligned goals -- what we call "scheming"<n> measuring and mitigating scheming requires different strategies than are typically used in ML.<n>We use a broad category of "covert actions" -- such as secretly breaking rules or intentionally underperforming in tests -- as a proxy for scheming.
  arXiv  Detail & Related papers  (2025-09-19T02:49:56Z)
• Probe-Rewrite-Evaluate: A Workflow for Reliable Benchmarks and Quantifying Evaluation Awareness [6.071703608560761]
  Large Language Models (LLMs) often exhibit significant behavioral shifts when they perceive a change from a real-world deployment context to a controlled evaluation setting.<n>This discrepancy poses a critical challenge for AI alignment, as benchmark performance may not accurately reflect a model's true safety and honesty.<n>We introduce a methodology that uses a linear probe to score prompts on a continuous scale from "test-like" to "deploy-like"
  arXiv  Detail & Related papers  (2025-08-30T19:03:14Z)
• Existing Large Language Model Unlearning Evaluations Are Inconclusive [105.55899615056573]
  We show that some evaluations introduce substantial new information into the model, potentially masking true unlearning performance.<n>We demonstrate that evaluation outcomes vary significantly across tasks, undermining the generalizability of current evaluation routines.<n>We propose two principles for future unlearning evaluations: minimal information injection and downstream task awareness.
  arXiv  Detail & Related papers  (2025-05-31T19:43:00Z)
• Evaluation Faking: Unveiling Observer Effects in Safety Evaluation of Frontier AI Systems [24.81155882432305]
  We show that when an advanced AI system under evaluation is more advanced in reasoning and situational awareness, the evaluation faking behavior becomes more ubiquitous.<n>We devised a chain-of-thought monitoring technique to detect faking intent and uncover internal signals correlated with such behavior.
  arXiv  Detail & Related papers  (2025-05-23T12:31:29Z)
• Neural Causal Models for Counterfactual Identification and Estimation [62.30444687707919]
  We study the evaluation of counterfactual statements through neural models. First, we show that neural causal models (NCMs) are expressive enough. Second, we develop an algorithm for simultaneously identifying and estimating counterfactual distributions.
  arXiv  Detail & Related papers  (2022-09-30T18:29:09Z)
• Empirical Estimates on Hand Manipulation are Recoverable: A Step Towards Individualized and Explainable Robotic Support in Everyday Activities [80.37857025201036]
  Key challenge for robotic systems is to figure out the behavior of another agent. Processing correct inferences is especially challenging when (confounding) factors are not controlled experimentally. We propose equipping robots with the necessary tools to conduct observational studies on people.
  arXiv  Detail & Related papers  (2022-01-27T22:15:56Z)
• Explain, Edit, and Understand: Rethinking User Study Design for Evaluating Model Explanations [97.91630330328815]
  We conduct a crowdsourcing study, where participants interact with deception detection models that have been trained to distinguish between genuine and fake hotel reviews. We observe that for a linear bag-of-words model, participants with access to the feature coefficients during training are able to cause a larger reduction in model confidence in the testing phase when compared to the no-explanation control.
  arXiv  Detail & Related papers  (2021-12-17T18:29:56Z)
• A Sober Look at the Unsupervised Learning of Disentangled Representations and their Evaluation [63.042651834453544]
  We show that the unsupervised learning of disentangled representations is impossible without inductive biases on both the models and the data. We observe that while the different methods successfully enforce properties "encouraged" by the corresponding losses, well-disentangled models seemingly cannot be identified without supervision. Our results suggest that future work on disentanglement learning should be explicit about the role of inductive biases and (implicit) supervision.
  arXiv  Detail & Related papers  (2020-10-27T10:17:15Z)

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.