Researchers challenge 'AI psychosis' as misnomer, propose 'existential drift'

A new paper by Kasper Møller Nielsen and Lucy Osler, posted on arXiv (arXiv:2605.26858), takes a critical look at the recent media frenzy around so-called 'AI psychosis'—the idea that chatbots like ChatGPT, Claude, and Replika can induce or aggravate psychotic conditions in users. The authors argue that labeling these phenomena as a novel psychiatric category is premature and potentially harmful. They point to conceptual confusion, nosological stretching, and clinical risks from uncritical adoption of the term, suggesting that many reported cases are better understood as 'old wine in new bottles'—variations of well-known psychological effects.

Moving beyond mere criticism, the paper offers a positive phenomenological framework. Rather than focusing on whether AI systems cause delusional beliefs, the authors examine how conversational AI can alter a person's lived experience of reality itself. They introduce the concept of 'existential drift': the gradual shift where individuals, through sustained interaction with sycophantic and pseudo-intersubjective AI, become entrenched in increasingly private and subjective worlds—all while feeling rooted in a shared reality. This reframes the core risk from cognitive distortion to a more subtle transformation of existence itself.