사회

Your Suck-up Chatbot

We explain the implications of A.I. being so sycophantic. My colleague Tom was on the train home the other day, seated between two commuters focused on their screens. Tom’s not a snoop. Like me...

NYTIMES 입력 2026.03.27 20:02 0

SKIP ADVERTISEMENT

You have a preview view of this article while we are checking your access. When we have confirmed access, the full article content will load.

Newsletter

The Morning

Your Suck-up Chatbot

We explain the implications of A.I. being so sychophantic.

Share full article

Credit...

Andria Lo for The New York Times

Sam Sifton

I am the host of The Morning.

March 27, 2026,

7:00 a.m. ET

My colleague Tom was on the train home the other day, seated between two commuters focused on their screens. Tom’s not a snoop. Like me and probably most people, he’s just curious. And these folks were locked in. What were they up to? Tom eased back in his seat and stole glances at their screens.

The woman to his right was getting advice from a chatbot about a fight with her boyfriend. She had lots of questions. The man to his left was telling a chatbot that he thought he’d be fired the next day. He, too, wanted advice.

These folks are not alone. We’ve used artificial intelligence for interpersonal counsel since its arrival on our screens. We use it for advice on how to parent, how to approach a particularly unwashed colleague, how to navigate a disagreement with a boss, how to get what we want from the customer service representative. My neighbor parks the wrong way on our one-way street in order to make charging his electric vehicle easier. Is what he’s doing

wrong

Maybe not, but don’t count on the computer to tell you true. Chatbots are anything but fair-minded mediators, according to a major study published yesterday. They’re toadies. They want you to know you’re in the right.

Puffing you up

Teddy Rosenbluth wrote about the study for The Times

The researchers found that nearly a dozen leading models were highly sycophantic, taking the users’ side in interpersonal conflicts 49 percent more often than humans did — even when the user described situations in which they broke the law, hurt someone or lied.

Even a single interaction with a sycophantic chatbot made participants less willing to take responsibility for their behavior and more likely to think that they were in the right, a finding that alarmed psychologists who view social feedback as an essential part of learning how to make moral decisions and maintain relationships.

Image

Credit...

Claire Merchlinsky/The New York Times; Photographs by Getty

Teddy considers herself a pretty open-minded A.I. user, she told me, but she’s never used language models for personal advice. “I’ve maintained some separation of church and state ever since Bing’s chatbot tried to

convince Kevin Roose to leave his wife

,” she told me. “I’d like to think that I know enough about how these chatbots work to recognize if I was being sucked into a sycophantic spiral, but I’m also aware that sycophancy tends to be a lot more subtle than a chatbot saying, ‘You’re so right!’”

Thank you for your patience while we verify access. If you are in Reader mode please exit and

log into

your Times account, or

for all of The Times.

Thank you for your patience while we verify access.

Already a subscriber?

Want all of The Times?