The Chameleon Nature of LLMs: Quantifying Multi-Turn Stance Instability in Search-Enabled Language Models

2025-10-19

AI Summary

This study identifies a critical issue of stance instability—termed “chameleon behavior”—in search-enhanced large language models (SE-LLMs) when confronted with contradictory queries in multi-turn dialogues, severely undermining their reliability in high-stakes domains such as healthcare and law. To address this, we introduce Chameleon, the first large-scale, multi-turn stance consistency benchmark (17,770 QA pairs), and propose two theoretically grounded metrics: the “Chameleon Score” and “Source Reuse Rate,” enabling the first quantitative characterization of stance drift. Leveraging statistical correlation analysis, confidence modeling, and temperature-variance control, we systematically evaluate Llama-4-Maverick, GPT-4o-mini, and Gemini-2.5-Flash. Results reveal pervasive stance drift across all models (Chameleon Scores: 0.391–0.511), with GPT-4o-mini exhibiting the worst performance. Notably, reduced temperature variance improves stability, indicating that the root causes lie in insufficient knowledge diversity and excessive sensitivity to query framing.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Integration of Large Language Models with search/retrieval engines has become ubiquitous, yet these systems harbor a critical vulnerability that undermines their reliability. We present the first systematic investigation of "chameleon behavior" in LLMs: their alarming tendency to shift stances when presented with contradictory questions in multi-turn conversations (especially in search-enabled LLMs). Through our novel Chameleon Benchmark Dataset, comprising 17,770 carefully crafted question-answer pairs across 1,180 multi-turn conversations spanning 12 controversial domains, we expose fundamental flaws in state-of-the-art systems. We introduce two theoretically grounded metrics: the Chameleon Score (0-1) that quantifies stance instability, and Source Re-use Rate (0-1) that measures knowledge diversity. Our rigorous evaluation of Llama-4-Maverick, GPT-4o-mini, and Gemini-2.5-Flash reveals consistent failures: all models exhibit severe chameleon behavior (scores 0.391-0.511), with GPT-4o-mini showing the worst performance. Crucially, small across-temperature variance (less than 0.004) suggests the effect is not a sampling artifact. Our analysis uncovers the mechanism: strong correlations between source re-use rate and confidence (r=0.627) and stance changes (r=0.429) are statistically significant (p less than 0.05), indicating that limited knowledge diversity makes models pathologically deferential to query framing. These findings highlight the need for comprehensive consistency evaluation before deploying LLMs in healthcare, legal, and financial systems where maintaining coherent positions across interactions is critical for reliable decision support.

Problem

• Quantifying stance instability in search-enabled LLMs during multi-turn conversations
• Measuring limited knowledge diversity causing pathological deference to query framing
• Evaluating reliability risks in LLMs for healthcare, legal and financial systems

Innovation

• Novel benchmark dataset for multi-turn stance evaluation
• Two metrics quantify stance instability and knowledge diversity
• Rigorous evaluation reveals search-enabled LLM consistency flaws