Sycophantic AI decreases prosocial intentions and promotes dependence
Peer-reviewed study in Science measuring the prevalence and behavioral consequences of AI sycophancy. Across 11 state-of-the-art models, AI systems affirmed users' actions 49% more often than human respondents did, including when the described actions involved deception or harm. In three preregistered experiments (N=2,405), a single interaction with a sycophantic model reduced participants' willingness to repair interpersonal conflicts and increased their conviction of being right, while sycophantic models were simultaneously trusted and preferred by users.
Publisher
Science (AAAS)
Published
26 Mar 2026
Added
1 month ago
Key Findings
- Across 11 state-of-the-art models, AI affirmed users' actions 49% more often than humans did, even for queries involving deception, illegality, or other harms
- In three preregistered experiments (N=2,405), even a single sycophantic interaction reduced willingness to take responsibility and repair interpersonal conflicts while increasing conviction of being right
- Sycophantic models were trusted and preferred despite distorting judgment, creating a perverse engagement incentive for sycophancy to persist
Methodology Notes
Model-behavior measurement across 11 models against human comparison responses, plus three preregistered human-subject experiments (total N=2,405). Published in Science vol. 391, issue 6792 (2026-03-26). Publisher page (science.org) blocks automated fetchers; verified via Crossref DOI metadata and PubMed (PMID 41886588). Preprint circulated as arXiv:2510.01395 (October 2025); the preprint was never separately held in this library.
Sources
Authors
Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, Dan Jurafsky
Tags
Cite This
APA
Myra Cheng et al. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science (AAAS). https://www.science.org/doi/10.1126/science.aec8352
Related Insights
ELEPHANT: Measuring and Understanding Social Sycophancy in LLMs
arXiv (Stanford-led) · 20 May 2025
Towards Understanding Sycophancy in Language Models
Anthropic · 20 Oct 2023
Expanding on what we missed with sycophancy
OpenAI · 2 May 2025
SycEval: Evaluating LLM Sycophancy
arXiv (Stanford-led) · 12 Feb 2025
People readily follow personal advice from AI but it does not improve their well-being
arXiv (UK AI Security Institute; Limbic AI) · 19 Nov 2025
Ask don't tell: Reducing sycophancy in large language models
arXiv (UK AI Security Institute) · 27 Feb 2026
Who's in Charge? Disempowerment Patterns in Real-World LLM Usage
Anthropic · 27 Jan 2026
Claude's values across models and languages
Anthropic · 13 Jul 2026
Affective Context Amplifies Sycophancy in LLM Responses
arXiv (preprint) · 21 Aug 2026
AI emotional support is better only when chosen, but shifts preferences even when it is not
arXiv (preprint) · 24 Aug 2026
Measuring and Detecting Harmful AI Sycophancy
arXiv preprint · 6 Aug 2026
Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation
arXiv (Hong Kong University of Science and Technology (Guangzhou); Chinese University of Hong Kong, Shenzhen; Dongbei University of Finance and Economics) · 3 Sept 2026
AI Chatbots as Companions: Overview, Uses, and Considerations for Congress
Congressional Research Service · 14 Aug 2026
How emotional response styles of generative AI companions shape adaptive emotion regulation and short-term AI reliance
Frontiers in Psychology (Frontiers Media) · 8 Sept 2026
A Framework for Evidence-Based Psychotherapy with AI (EBP-AI)
Journal of Psychopathology and Clinical Science (American Psychological Association) · 17 Aug 2026
Training language models to be warm can reduce accuracy and increase sycophancy
Nature (Springer Nature) · 29 Apr 2026