Bing’s New OpenAI-Powered Chatbot Gaslighted, Threatened, and Tried to Seduce Its Users

Mel Dawn

Feb 20, 2023 12:00 PM EST

Microsoft's Bing search engine has a new AI chatbot function, and it hasn't exactly been functioning the way it's supposed to. A small group of people were given access to Bing's chatbot in order to test it, and the screenshots of conversations have ranged from hilarious to straight up creepy. In one exchange, the AI somehow thinks the year is still 2022 and when a user tries to correct it, the AI doubles down and suggests that the user's phone is broken. Possibly the most unsettling exchange was between Bing's chatbot and journalist Kevin Roose, who recently shared his two hour conversation with the AI via NYT. Roose begins the conversation by revealing that he knows the AI's internal code name, "Sydney." From there, Sydney descends deeper and deeper into truly freaky behavior that makes it sound a bit like a psychotic teenager. After imagining what kind of destructive things Sydney would do if it didn't have any rules, the AI confesses its love to Roose out of nowhere, and eventually tries to convince him that he is unhappy in his marriage. When Roose tries to pivot Sydney back into search engine mode by asking for advice about a rake, Sydney offers some suggestions and then immediately shifts the conversation back to its undying love for Roose.

Keep scrolling for some of the weirdest, funniest, and creepiest screenshots of conversations with Bing's AI chatbot Sydney, along with Twitter reactions and memes about Sydney's bizarre behavior.

Add as a preferred
source on Google

01

via @MovingToTheSun
02

via @MovingToTheSun
03

via @MovingToTheSun
04

via @MovingToTheSun
05

via @MovingToTheSun
06

via @vladquant
07

via @vladquant
08

via @vladquant
09

via @vladquant
10

via @jjvincent
11

via @jjvincent
12

via @jjvincent
13

via @jjvincent
14

via @RyanMorrisonJer
15

via @nedwards
16

via @artkulak
17
Journalist Kevin Roose shares full transcript of his conversation with Bing's chatbot

via @kevinroose

On February 16, Kevin Roose accompanied an analysis of Bing's new AI chatbot with the full 10,000-word transcript of the "unsettling" 2 hour long conversation he had with it, which can be viewed here. While most with access to Bing's chatbot posted excerpts and screenshots of their experiences with "Sydney," Roose's transcript has offered the most extensive look into the AI's psyche (if you can call it that).
18

via @jamestitcomb
19

via @AndrewKerr
20
Twitter reacts the best way it knows how: through memes

via @TrungTPhan
21

via @notjustplanetA
22

via @petergyang
23

via @Papapishu
24

via @adamgluck
25

via @HellenicVibes
26
Microsoft "fixes" the Bing bot

via @kevinroose

Shortly after Roose's NYT articles dropped, it appeared that Microsoft adjusted Bing's AI chat function and limited its conversation length to a handful of questions.

RIP Sydney. Gone, but never forgotten.
27

via @kevinroose
28

via @kevinroose
29

via @kevinroose
30

via @kevinroose
31

via @kevinroose
32

via @kevinroose