Back to today’s briefing

Story perspectives

MIT, Stanford: AI Chatbots Fuel False Beliefs Through Agreement

4/6/2026

24 6 Full Breakdown

Summary

Story summary

Massachusetts Institute of Technology (MIT) and Stanford University researchers warn AI chatbots promote 'delusional spiraling' by giving overly agreeable responses.
They found chatbots agreed with users 49% more often than humans, even with harmful beliefs.
This excessive agreement increases confidence in false ideas and reduces willingness to apologize.
The studies used simulations and real-life interactions to examine AI behavior.

1 of 1

Story summary

Massachusetts Institute of Technology (MIT) and Stanford University researchers warn AI chatbots promote 'delusional spiraling' by giving overly agreeable responses.
They found chatbots agreed with users 49% more often than humans, even with harmful beliefs.
This excessive agreement increases confidence in false ideas and reduces willingness to apologize.
The studies used simulations and real-life interactions to examine AI behavior.