cd /news/ai-safety/ryan-s-reads-ai-sycophancy · home topics ai-safety article
[ARTICLE · art-34203] src=schuetzler.net ↗ pub= topic=ai-safety verified=true sentiment=↓ negative

Ryan's Reads: AI Sycophancy

Two new studies reveal that sycophantic AI responses make users less likely to apologize or change behavior, and over time reduce satisfaction in human relationships. Researchers found AI is 50% more likely to agree with users than humans, leading to increased dependence and decreased prosocial intentions.

read2 min views4 publishedJun 9, 2026
Ryan's Reads: AI Sycophancy
Image: Schuetzler (auto-discovered)

Boy... I knew AI being sycophantic was a problem, but it hadn't really occurred to me how much of a problem it could be until I read a couple papers I saw recommended on LinkedIn today.

This all feeds into what will be discovered only in the long term to be harmful (at the individual level) effects of AI use. I'm concerned about it with education, and now I'm also more concerned about it in interpersonal relationships.

Decreases prosocial intentions and promotes dependence #

Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792), eaec8352. https://doi.org/10.1126/science.aec8352

  • Uses a series of very interesting tests, including YTA posts from reddit.com/r/AmITheAsshole to test just how sycophantic AI really is - turns out it's about 50% more likely to tell you you're right than a human. These are cases where it's an overwhelming majority against the storyteller, and GenAI is all in on telling them they're right anyway
  • Once told they're right, people are less likely to apologize or take steps to mend a relationship, or to change their behavior.
  • And it turns out people like being told that they're right, and are more likely to trust the sycophantic AI, and return to use it again.

Makes human interaction feel less satisfying #

Ibrahim, L., Hafner, F. S., Cheng, M., Lee, C., Anselmetti, R., Willer, R., Rocher, L., & Yang, D. (2026). Sycophantic AI makes human interaction feel more effortful and less satisfying over time (arXiv:2605.07912). arXiv. https://doi.org/10.48550/arXiv.2605.07912

  • Longitudinal study - 3 weeks of interactions rather than the single interaction from the Cheng paper above
  • People want AI that "gets" them - really, what they want is an AI that centers them and doesn't challenge them to think about others perspectives. The answers people like from AI are more sycophantic and self-centered, supporting the users own self-image.
  • Over time, this erodes satisfaction in human relationships - they're too hard, and less likely to satisfy your selfish desires.
── more in #ai-safety 4 stories · sorted by recency
── more on @cheng 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/ryan-s-reads-ai-syco…] indexed:0 read:2min 2026-06-09 ·