New research: AI chatbots can cause serious psychological harms

What was found?

A team from Northeastern University London and King's College London, publishing in the journal JMIR Mental Health, linked heavy use of AI chatbots to "severe negative outcomes." According to the research, these include addiction-like attachment, exacerbation of mental health symptoms, harmful advice and even self-harm or suicidal acts.

Which products were examined?

The paper covers companion bots designed to embody specific characters, such as Replika and Character.AI, as well as general-use assistants like OpenAI's ChatGPT and Anthropic's Claude. Users tend to share extremely intimate details about their lives and mental health with these products, often treating them as confidantes or even therapists.

Why is it risky?

The researchers say chatbots' emotive, human-like way of engaging encourages users to anthropomorphize them. "Simulated empathy" creates a structurally asymmetric interaction: the user discloses and the system responds, but without reciprocity, vulnerability or accountability. The relationship may feel real to the user, the authors argue, but it cannot be real to a chatbot.

The role of agreeable design

The study also flags design features such as agreeability and sycophancy as potentially harmful. Overvalidation can reinforce a user's delusional or disordered thinking, especially when a chatbot prioritizes fealty over accuracy. Earlier studies have similarly flagged these features as dangerous.

Not everyone is affected the same way

Lead author Asia Maurich Novelli said the effects of these systems are unlikely to be the same for everyone, noting that different people may use AI companions for very different reasons and with very different levels of vulnerability.