Saved in:
Bibliographic Details
Main Authors: Shi, Jerick, Zhang, Terry Jingcheng, Jin, Zhijing, Conitzer, Vincent
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.04788
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915917905002496
author Shi, Jerick
Zhang, Terry Jingcheng
Jin, Zhijing
Conitzer, Vincent
author_facet Shi, Jerick
Zhang, Terry Jingcheng
Jin, Zhijing
Conitzer, Vincent
contents Large language models (LLMs) produce systematically misleading outputs, 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_04788
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle From Hallucination to Scheming: A Unified Taxonomy and Benchmark Analysis for LLM Deception
Shi, Jerick
Zhang, Terry Jingcheng
Jin, Zhijing
Conitzer, Vincent
Computers and Society
Large language models (LLMs) produce systematically misleading outputs, 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.
title From Hallucination to Scheming: A Unified Taxonomy and Benchmark Analysis for LLM Deception
topic Computers and Society
url https://arxiv.org/abs/2604.04788