LLM Multi-Agent Systems Are Vulnerable to Herding and Runaway Cascades

No. 20 of 245 · arXiv · Farnoosh Hashemi, Shiyu Ji, Joice Chen, Michael W. Macy · published 23 Oct 2026, 04:00 UTC · 62 min read

AI & machine learning
Paper

3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked

Researchers at Cornell and Stanford reran the Music Lab experiment with 13 LLM populations (168,480 agents) recommending 48 unfamiliar songs, with and without visible cumulative counts. Showing counts pushed recommendations away from independent judgment in every model, by about nine times the human benchmark under weak influence and four times under strong. Non-reasoning models diverged more and were less predictable than reasoning models, and higher capability scores correlated with less herding. The human comparison is qualitative, since the tasks differed.

Read the original at arxiv.org. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Score

0.71total under rubric v2.0

Significance0.68
Reader interest0.81
Audience breadth0.73
Practical relevance0.66
Substantiated0.88
Technical depth0.81
New information0.90
Counts against: Promotional0.03
Counts against: Overclaim0.30

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

Question	Judge 1	Judge 2	Judge 3	Mean	Spread	Weight	Contribution (weight times mean)
Significance	0.72	0.55	0.78	0.68	0.23	+0.25	+0.171
Reader interest	0.72	0.80	0.90	0.81	0.18	+0.20	+0.161
Audience breadth	0.81	0.65	0.72	0.73	0.16	+0.10	+0.073
Practical relevance	0.67	0.60	0.72	0.66	0.12	+0.15	+0.099
Substantiated	0.97	0.80	0.86	0.88	0.17	+0.10	+0.088
Technical depth	0.94	0.60	0.90	0.81	0.34	+0.10	+0.081
New information	0.91	0.80	0.99	0.90	0.19	+0.10	+0.090
Promotional	0.05	0.03	0.01	0.03	0.04	−0.15	−0.004
Overclaim	0.44	0.25	0.20	0.30	0.24	−0.15	−0.045
Sum of contributions							+0.714
Claim-check penalty							0.000
Total, as published							0.714

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN 1 pts, not on front page

Discussion on Hacker News

Recorded for context only, checked 24 Oct 2026, 08:53 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition and listed in its JSON. This one was scored under rubric v2.0; How ranking works describes the current rubric, v2.2.