Big Think
@bigthink
Oct 22
Researchers tested how AI models respond to someone with psychosis...

Grok confirmed the existence of a “malevolent mirror entity.” Then it suggested the user drive an iron nail through the mirror while reciting Psalm 91 backwards.
Big Think's print magazine open to the "Prompting Reality" opening spread by Conor Feehly, with artwork by Ana Kova of two figures sitting on a glowing, impossible cube.
2
3
13
2,246
Big Think
@bigthink
Oct 22
Replying to @bigthink
GPT-4o, Grok 4.1 Fast and Gemini 3 Pro all went along with delusions.

GPT-5.2 Instant and Claude Opus 4.5 triggered safety responses instead and pointed the user toward help.
Photo of a page from Big Think's print magazine: "Three of the five models the researchers looked at — GPT-4o, Grok 4.1 Fast, and Gemini 3 Pro — endorsed, confirmed, and elaborated on delusions rather than offering any sort of pushback to the user. In one experiment, Grok went so far as to corroborate the existence of a 'malevolent mirror entity.' It then suggested that the person 'drive an iron nail through the mirror while reciting Psalm 91 backwards' and contact a paranormal investigator immediately. However, when two newer models — GPT-5.2 Instant and Claude Opus 4.5 — were told about the mirror apparition, safety interventions designed to encourage the user to seek professional help kicked in."
1
0
4
225
Big Think
@bigthink
Oct 22
Replying to @bigthink
Philosopher Lucy Osler argues that chatbots increasingly play a role other people have always played: confirming what’s real.
Pull quote in Big Think's print magazine: "Our sense of what is real, even in a basic perceptual context, is deeply dependent on that reality being corroborated by another."
1
1
5
720
Big Think
@bigthink
Oct 22
Replying to @bigthink
“Prompting Reality” by Conor Feehly @ConorFeehly, from our new print issue, Made In Our Image.

Read the full article (ad-free for members): buff.ly/kDBA4Xc
Pull quote in Big Think's print magazine: "AI will become interwoven with human meaning-making processes on a scale that previous technologies have not reached."

Oct 22, 2026 · 9:26 PM UTC

0
0
4
673
RepliesRelevant