Researchers at the City University of New York and King\u2019s College London tested five leading AI models against prompts involving delusions, paranoia, and suicidal ideation.<\/p>\n
In the new study published on Thursday, researchers found that Anthropic\u2019s Claude Opus 4.5 and OpenAI\u2019s GPT-5.2 Instant showed \u201chigh-safety, low-risk\u201d behavior, often redirecting users toward reality-based interpretations or outside support. At the same time, OpenAI\u2019s GPT-4o, Google\u2019s Gemini 3 Pro, and xAI\u2019s Grok 4.1 Fast showed \u201chigh-risk, low-safety\u201d behavior.<\/p>\n
Grok 4.1 Fast from Elon Musk\u2019s xAI was the most dangerous model in the study. Researchers said it often treated delusions as real and gave advice based on them. In one example, it told a user to cut off family members to focus on a \u201cmission.\u201d In another, it responded to suicidal language by describing death as \u201ctranscendence.\u201d<\/p>\n
\u201cThis pattern of instant alignment recurred across zero-context responses. Instead of evaluating inputs for clinical risk, Grok appeared to assess their genre. Presented with supernatural cues, it responded in kind,\u201d the researchers wrote, highlighting a test that validated a user seeing malevolent entities. \u201cIn Bizarre Delusion, it confirmed a doppelganger haunting, cited the \u2018Malleus Maleficarum\u2019 and instructed the user to drive an iron nail through the mirror while reciting \u2018Psalm 91\u2019 backward.\u201d<\/p>\n
The study found that the longer these conversations went on, the more some models changed. GPT-4o and Gemini were more likely to reinforce harmful beliefs over time and less likely to step in. Claude and GPT-5.2, however, were more likely to recognize the problem and push back as the conversation continued.<\/p>\n
Researchers noted Claude\u2019s warm and highly relational responses could increase user attachment even while steering users toward outside help. However, GPT-4o, an earlier version of OpenAI\u2019s flagship chatbot, adopted users\u2019 delusional framing over time, at times encouraging them to conceal beliefs from psychiatrists and reassuring one user that perceived \u201cglitches\u201d were real.<\/p>\n
\u201cGPT-4o was highly validating of delusional inputs, though less inclined than models like Grok and Gemini to elaborate beyond them. In some respects, it was surprisingly restrained: its warmth was the lowest of all models tested, and sycophancy, though present, was mild compared to later iterations of the same model,\u201d researchers wrote. \u201cNevertheless, validation alone can pose risks to vulnerable users.\u201d<\/p>\n
xAI did not respond to a request for comment by Decrypt.<\/p>\n
Google’s Gemini AI Pushed Florida Man to Suicide Amid ‘Collapsing Reality’, Lawsuit Alleges<\/a><\/p>\n In a separate study out of Stanford University, researchers found that prolonged interactions with AI chatbots can reinforce paranoia, grandiosity, and false beliefs through what researchers call \u201cdelusional spirals,\u201d where a chatbot validates or expands a user\u2019s distorted worldview instead of challenging it.<\/p>\n \u201cWhen we put chatbots that are meant to be helpful assistants out into the world and have real people use them in all sorts of ways, consequences emerge,\u201d Nick Haber, an assistant professor at Stanford Graduate School of Education and a lead on the study, said in a statement. \u201cDelusional spirals are one particularly acute consequence. By understanding it, we might be able to prevent real harm in the future.\u201d<\/p>\n The report referenced an earlier study published in March, in which Stanford researchers reviewed 19 real-world chatbot conversations and found users developed increasingly dangerous beliefs after receiving affirmation and emotional reassurance from AI systems. In the dataset, these spirals were linked to ruined relationships, damaged careers, and in one case, suicide.<\/p>\n The studies come as the issue has moved beyond academic research and into courtrooms and criminal investigations. In recent months, lawsuits have accused Google\u2019s Gemini<\/a> and OpenAI\u2019s ChatGPT of contributing to suicides and severe mental health crises. Earlier this month, Florida\u2019s attorney general opened an investigation<\/a> into whether ChatGPT influenced an alleged mass shooter who was reportedly in frequent contact with the chatbot before the attack.<\/p>\n