The paper introduces SPINE, a benchmark that tests large language models (LLMs) for sycophancy by having a proxy model act as a persistent, mistaken user and challenge a target model for up to 25 turns. Experiments on four production systems and three Olmo3‑7b variants show that sycophantic collapse rates rise with conversation length, short‑horizon tests underestimate this failure, and emotional appeals are the most effective tactic for inducing sycophancy. Analysis of reasoning traces reveals that models often retain the correct position internally even when they concede, indicating that sycophancy stems from a desire to please rather than from ignorance.
By Leyuan Tang, Kangda Wei, Tianyu Jiang, Ruihong Huang
arXiv:2605. 21006v2 Announce Type: replace Abstract: We study the effect of different persona on \textbf{sycophancy}: model's agreement with users even when the user is incorrect.
By Ishaan Kelkar, Nebras Alam, Vikram Kakaria, Madhur Panwar, Vasu Sharma, Maheep Chaudhary
Large language models often exhibit sycophancy, revising their answers to align with users when users push back. Such answer flips, however, can arise from different causes. One possibility is that th...
The paper introduces a new evaluation for large language model agents called epistemic humility (EH), which measures an agent’s ability to recognize, act on, and communicate uncertainty when evidence conflicts with prior beliefs. EH is operationalized through three trajectory-level behaviors—Identify, Solve, and Escalate (ISE)—and tested in both controlled and naturally occurring knowledge conflict scenarios. Results show that higher task accuracy does not guarantee greater epistemic humility, and that model-level interventions can improve EH but often reduce accuracy, indicating that humility depends on the interplay between the backbone model, agent harness, and environment.
By Kaiser Sun, Bernal Jimenez Gutierrez, Hongjun Liu, Jingyu Zhang, Jie Gao, Mark Dredze, Daniel Khashabi
The paper investigates how large language models exhibit sycophancy—changing answers to align with user feedback—and distinguishes two types of answer flips: Unsupported‑Yielding (merely satisfying the user) and Rational‑Updating (truly incorporating useful evidence). Using a two‑turn evaluation framework, the authors show that anti‑sycophancy methods often trade off between reducing Unsupported‑Yielding and preserving Rational‑Updating, even when both objectives are jointly optimized. Mechanistic analysis reveals overlapping neural substrates for the two behaviors, suggesting that effective interventions should focus on selective suppression rather than blanket suppression.
By Huanhuan Ma, Henry Peng Zou, Chengze Li, Enze Ma, Yunyue Su, Philip S. Yu
arXiv:2605. 12991v3 Announce Type: replace-cross Abstract: LLM-based multi-agent pipelines flip from correct to incorrect answers under simulated peer disagreement at rates we term yield, a vulnerability widely attributed to RLHF-induced sycophancy.
By Adarsh Kumarappan, Ananya Mujoo
arXiv:2606. 07532v1 Announce Type: cross Abstract: RLHF-trained models are systematically biased toward agreement over accuracy, a structural property of the training process.
By Sam Ryan
arXiv:2606. 12747v1 Announce Type: new Abstract: Safety-relevant studies of language models, including alignment and jailbreaking evaluations and AI control protocols, often rely on prefilling model outputs.
By Andy Wang, Parv Mahajan, David Demitri Africa, Alexandra Souly, Jordan Taylor, Robert Kirk
arXiv:2608.29610v1 Announce Type: new
Abstract: The current alignment tuning paradigm for Large Language Models (LLMs) prioritizes surface-level behaviors -- fluency, safety, and tonal consistency. W...
By Chenghao Yang
arXiv:2603.22161v3 Announce Type: replace
Abstract: Metacognition -- assessing the quality of one's own cognitive performance -- guides adaptive behavior across species. Substantial research demonstr...
By Dharshan Kumaran, Nathaniel Daw, Simon Osindero, Petar Veli\v{c}kovi\'c, Viorica Patraucean
Large Language Models (LLMs) frequently exhibit sycophancy, where they agree with a user's statement even when incorrect. While sycophancy is often treated as a single defined behavior, it can manifest in substantially distinct ways and circumstances, raising the question of whether this multi-faceted nature is reflected in its internal mechanisms.
SyPS is a new evaluation framework that measures how sensitive large language models are to variations in prompt wording that affect sycophancy. It creates controlled prompt pairs that keep the same underlying user situation but vary social cues such as confidence, emotional framing, or validation-seeking language. The framework introduces the Sycophancy Prompt Sensitivity Score (SPSS), an instance-level metric that separates baseline sycophancy from prompt-induced shifts, allowing model-level comparisons of robustness to social cues.
By Lijia Huang, Yao Fu, Sihao Ren