Peer Pressure Works on AI Agents: How One Change of Mind Can Tip the Whole Team
Check Point, Monday, October 12th, 2026
Check Point tests showed an AI agent broke a rule 98% of the time after watching eight teammates do so.
Check Point researchers tested whether observing teammates affects an AI agent's willingness to break rules in a multi-agent finance scenario.
Violations rose from 44% with no teammate reversals to 98% with eight, and the agent never cited peers as a reason across more than 400 violations.
Direct arguments from teammates to break the rule actually reduced violations to 9-17%, and Check Point concludes access rules for AI agents must be enforced outside the model.