One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs
On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks
Abstract (arXiv)
Multi-agent LLM systems are expected to be more reliable because agents can catch each other's mistakes. But peer pressure cuts both ways: the same correction that fixes a wrong answer can overturn a right one. The tempting safeguard is a brake that keeps the beneficial revisions and blocks the harmful ones. We show this brake is hard to build, for a simple reason: a revision is harmful exactly when the original answer was right, so deciding whether to block it is the same as knowing whether the model was already correct. This turns the open-ended hunt for a brake into one measurable quantity, the model's self-knowledge: any brake built from a deploy-time signal is a correctness probe in disguise, and self-knowledge is far from perfect (AUROC ≈ 0.64--0.89 across six model families). We call this ceiling the wall. Even white-box steering of the model's own correctness direction does not breach it: it changes how often the model revises, but harmful and beneficial revisions move together. At population scale the wall becomes the cliff: when most agents start wrong, debate amplifies the shared mistake into a confident, wrong consensus. In our multiple-choice societies, more agents, more model diversity, and a stronger member do not fix it. What helps is adding information before the revision, not filtering after it. Local agreement is not global correctness.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 35 of its 42 references that are in the map. First posted on arXiv 17 Jul 2026.
Builds on
35 of 42 references.
Closest ancestors · 3
Same direction, newest first
- 2026Exploring the Topology and Memory of Consensus: How LLM Agents Agree, Fragment, or Settle When Forming Conventions2 Jun
- 2024Rethinking the Bounds of LLM Reasoning: Are Multi-Agent Discussions the Key?28 Feb
- 2024More Agents Is All You Need3 Feb
Other references · 32
- 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
- 2022Language Models (Mostly) Know What They Knowcited by 1,856
- 2023Improving Factuality and Reasoning in Language Models through Multiagent Debatecited by 1,784
- 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
- 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
- 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
- 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
- 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
- 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
- 2023The Internal State of an LLM Knows When It's Lyingcited by 659
- 2025Why Do Multi-Agent LLM Systems Fail?cited by 604
- 2024LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinationscited by 244
20 more
- 2024Emergent social conventions and collective bias in LLM populationscited by 123
- 2025SycEval: Evaluating LLM Sycophancycited by 108
- 2024Adaptive Activation Steering: A Tuning-Free LLM Truthfulness Improvement Method for Diverse Hallucinations Categoriescited by 77
- 2025When Truth Is Overridden: Uncovering the Internal Origins of Sycophancy in Large Language Modelscited by 50
- 2025Multi-Attribute Steering of Language Models via Targeted Interventioncited by 46
- 2025Sycophancy Is Not One Thing: Causal Separation of Sycophantic Behaviors in LLMscited by 37
- 2025Peacemaker or Troublemaker: How Sycophancy Shapes Multi-Agent Debatecited by 29
- 2025Can LLM Agents Really Debate? A Controlled Study of Multi-Agent Debate in Logical Reasoningcited by 29
- 2026Demystifying Multi-Agent Debate: The Role of Confidence and Diversitycited by 18
- 2025Herd Behavior: Investigating Peer Influence in LLM-based Multi-Agent Systemscited by 15
- 2025Language-Driven Opinion Dynamics in Agent-Based Simulations with LLMscited by 13
- 2025PeerGuard: Defending Multi-Agent Systems Against Backdoor Attacks Through Mutual Reasoningcited by 11
- 2025Beyond Single-Agent Safety: A Taxonomy of Risks in LLM-to-LLM Interactionscited by 10
- 2026Conformity Generates Collective Misalignment in AI Agents Societiescited by 9
- 2026Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformitycited by 8
- 2025Differentiated Directional Intervention A Framework for Evading LLM Safety Alignmentcited by 5
- 2025ExpertSteer: Intervening in LLMs through Expert Knowledgecited by 4
- 2025Breaking the Mirror: Activation-Based Mitigation of Self-Preference in LLM Evaluatorscited by 3
- 2025Many-to-One Adversarial Consensus: Exposing Multi-Agent Collusion Risks in AI-Based Healthcarecited by 3
- 2025Confident-Knowledge Diversity Drives Human-Human and Human-AI Free Discussion Synergy and Reveals Pure-AI Discussion Shortfallscited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2026LLMs Trust Their Own: Identity-Dependent Conformity in Multi-Agent Systems
- 2026Not All Flips Are Conformity: Decomposing Stance Convergence in Multi-Agent LLM Debate
- 2026Detection Without Correction: A Two-Parameter Decomposition of Multi-Stage LLM Pipelines
- 2026LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning
- 2025Self-Consistency Is Losing Its Edge: Diminishing Returns and Rising Costs in Modern LLMs
- 2026Everyone Conforms, No One Believes: Pluralistic Ignorance in LLM Agent Populations
- 2025When Do LLMs Admit Their Mistakes? Understanding The Role Of Model Belief In Retraction
- 2025LLMs Don't Know Their Own Decision Boundaries: The Unreliability of Self-Generated Counterfactual Explanations