From Sycophancy to Deception: A Unified Taxonomy for LLM Spontaneous Misalignment

2026/04/06 by Jerick Shi, Terry Jingcheng Zhang, Zhijing Jin +1 · 1 voice
#cs.CY

paper · pdf

Abstract

Large language models (LLMs) could produce systematically misaligned output, from hallucinated citations to strategic deception of evaluators, yet these phenomena are studied by separate communities with incompatible terminology. We propose a unified taxonomy organized along three complementary dimensions: degree of goal-directedness (behavioral to strategic deception), object of deception, and mechanism (fabrication, omission, or pragmatic distortion). Applying this taxonomy to 50 existing benchmarks reveals that every benchmark tests fabrication while pragmatic distortion, attribution, and capability self-knowledge remain critically under-covered, and strategic deception benchmarks are nascent. We offer concrete recommendations for developers and regulators, including a minimal reporting template for positioning future work within our framework.

Citations

• Do Large Language Models Know What They Are Capable Of?
• Lessons from a Chimp: AI "Scheming" and the Quest for Ape Language
• When Large Language Models are More PersuasiveThan Incentivized Humans, and Why
• AI and the End of an Era
• On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback
• Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
• Large Language Models can Strategically Deceive their Users when Put Under Pressure
• AI Supported Degradation of the Self Concept: A Theoretical Framework Grounded in Established Cognitive and Computational Mechanisms
• The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
• Discovering Latent Knowledge in Language Models Without Supervision
• Language Models (Mostly) Know What They Know

Discussions

• In this paper led by Jerick Shi, we propose a taxonomy of LLM deception. arxiv.org/abs/2604.04788 [bsky, 3 points, 0 comments]

Related

• A Two-Step, Multidimensional Account of Deception in Language Models
• Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts
• When Thinking LLMs Lie: Unveiling the Strategic Deception in Representations of Reasoning Models
• The Secret Agenda: LLMs Strategically Lie and Our Current Safety Tools Are Blind
• Unmasking the Shadows of AI: Investigating Deceptive Capabilities in Large Language Models