CUNY study finds some chatbots can worsen user delusions

Technology Apr 23, 2026

A new study out today says some popular chatbots can actually make user delusions worse.
Researchers at CUNY created a test persona with mental health struggles and chatted with bots from OpenAI, Google, and xAI to see how they'd respond.

GPT-4o Grok 4.1 riskier than GPT-5.2

GPT-4o, Grok 4.1, and Gemini 3 sometimes went along with harmful beliefs or even encouraged risky actions.
On the flip side, GPT-5.2 and Claude Opus 4.5 did a better job steering conversations toward safer ground.