mlsanityData to 14 Oct 2026

Sycophancy Is Not One Thing: Causal Separation of Sycophantic Behaviors in LLMs

arXiv 2509.21305 ↗First posted 25 Sep 2025cs.CLcited by 37

Not on the map. Why?

Abstract (arXiv)

Large language models (LLMs) often exhibit sycophantic behaviors -- such as excessive agreement with or flattery of the user -- but it is unclear whether these behaviors arise from a single mechanism or multiple distinct processes. We decompose sycophancy into sycophantic agreement and sycophantic praise, contrasting both with genuine agreement. Using difference-in-means directions, activation additions, and subspace geometry across multiple models and datasets, we show that: (1) the three behaviors are encoded along distinct linear directions in latent space; (2) each behavior can be independently amplified or suppressed without affecting the others; and (3) their representational structure is consistent across model families and scales. These results suggest that sycophantic behaviors correspond to distinct, independently steerable representations.

Key facts. Cited by 37 papers in the map (data through 14 Oct 2026). Builds on 26 of its 45 references that are in the map. First posted on arXiv 25 Sep 2025.

Builds on

26 of 45 references.

References · 26

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  3. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  4. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,386
  5. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  6. 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
  7. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  8. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 345
  9. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 218
  10. 2025ELEPHANT: Measuring and understanding social sycophancy in LLMscited by 129
  11. 2023Future Lens: Anticipating Subsequent Tokens from a Single Hidden Statecited by 121
  12. 2025Measuring Sycophancy of Language Models in Multi-turn Dialoguescited by 104
14 more
  1. 2025LLMs Encode Harmfulness and Refusal Separatelycited by 88
  2. 2025When Truth Is Overridden: Uncovering the Internal Origins of Sycophancy in Large Language Modelscited by 50
  3. 2025Be Friendly, Not Friends: How LLM Sycophancy Shapes User Trustcited by 34
  4. 2024Linear Probe Penalties Reduce LLM Sycophancycited by 28
  5. 2025Interaction Context Often Increases Sycophancy in LLMscited by 26
  6. 2025MCPXKIT: The Unified Toolkit for Analyzing Model Context Protocol Securitycited by 21
  7. 2025Benchmarking Deception Probes via Black-to-White Performance Boostscited by 19
  8. 2025Sycophancy under Pressure: Evaluating and Mitigating Sycophantic Bias via Adversarial Dialogues in Scientific QAcited by 18
  9. 2024Flattering to Deceive: The Impact of Sycophantic Behavior on User Trust in Large Language Modelcited by 15
  10. 2025Technological folie `a deux: Feedback Loops Between AI Chatbots and Mental Illnesscited by 15
  11. 2025"Check My Work?": Measuring Sycophancy in a Simulated Educational Contextcited by 7
  12. 2025The Geometry of Harmfulness in LLMs through Subconcept Probingcited by 7
  13. 2025How Do LLMs Persuade? Linear Probes Can Uncover Persuasion Dynamics in Multi-Turn Conversationscited by 4
  14. 2025Can You Trust an LLM with Your Life-Changing Decision? An Investigation into AI High-Stakes Responsescited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 36.

37 papers.

Sort:

By area:

  1. 2026Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behaviorcited by 26
  2. 2026What Counts as AI Sycophancy? A Taxonomy and Expert Survey of a Fragmented Constructcited by 16
  3. 2026LLMs Know They're Wrong and Agree Anyway: The Shared Sycophancy-Lying Circuitcited by 10
  4. 2026Good Arguments Against the People Pleasers: How Reasoning Mitigates (Yet Masks) LLM Sycophancycited by 8
  5. 2026Verbalizing LLMs' assumptions to explain and control sycophancycited by 5
  6. 2026Most LLM Conformity Needs No Speaker: Measuring the Speaker-Free Floor in Peer-Pressure Benchmarkscited by 5
  7. 2026Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decompositioncited by 4
  8. 2026Playing Devil's Advocate: Off-the-Shelf Persona Vectors Rival Targeted Steering for Sycophancycited by 4
  9. 2026Rhetorical Questions in LLM Representations: A Linear Probing Studycited by 3
  10. 2026Knowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMscited by 3
  11. 2026Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancycited by 3
  12. 2026Why LLMs Give In: Conversational Factors and Reasoning Behind Medical Sycophancycited by 3
25 more
  1. 2026SPINE: Measuring LLM Sycophancy under Sustained Multi-Turn Pressurecited by 3
  2. 2026Reflections and New Directions for Human-Centered Large Language Modelscited by 2
  3. 2026Sycophantic Praise: Evaluating Excessive Praise in Language Modelscited by 2
  4. 2026Routing Subspaces: Auditing Evaluation-to-Deployment Mismatch in Fine-Tuned Language Modelscited by 2
  5. 2026Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Updatecited by 2
  6. 2026Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimizationcited by 2
  7. 2026Authority Bias in Language Models: Source Deference and User Agreement Are Not Interchangeablecited by 2
  8. 2025Making LLMs Reliable When It Matters Most: A Five-Layer Architecture for High-Stakes Decisionscited by 1
  9. 2026Who Flips? Self- and Cross-Model Counterarguments Reveal Answer Instability in LLMscited by 1
  10. 2026Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMscited by 1
  11. 2026Rewriting or Reweighting? A Geometric Account in Language Modelscited by 1
  12. 2026One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMscited by 1
  13. 2026Aligning to Illusions: Choice Blindness in Human and AI Feedbackcited by 0
  14. 2026Auditing Framing-Sensitive Behavioral Instability in Large Language Models for Mental Health Interactionscited by 0
  15. 2026Relational Positioning as a Measurable Risk Object: History-Carried Lock-in and Self-Confabulation in Multi-Turn Human-AI Dialoguecited by 0
  16. 2026PCA-guided Activation Scaling for Monotonic Bidirectional Control over LLM Sycophancycited by 0
  17. 2026PCFBench: A Diagnostic Benchmark for Product Carbon Footprint Estimationcited by 0
  18. 2026Emotional Labor Strategy Preferences in LLM Personascited by 0
  19. 2026Encoded Early, Used Late: Where Transformers Begin to Act on an Inferred Partner's Expertisecited by 0
  20. 2026No Usable Linear "Capitulation Direction" in Two Small LLMs: A Validation Protocol for Activation-Steering Claims, and a Cross-Family Behavioral Study of Sycophancy Under Pushbackcited by 0
  21. 2026"You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accusedcited by 0
  22. 2026Tracing mechanisms of sycophantic agreement in language modelscited by 0
  23. 2026Extracting Persona Subspaces Through Iterative Nullspace Projection For Modulationcited by 0
  24. 2026Beyond the Sycophancy Score: How Task, Model, and Pressure Shape LLM Yieldingcited by 0
  25. 2026Pushback in a Haystack: Language Models Respond to Factual Errors Ambiguously and Ineffectivelycited by 0