Large language models (LLMs) can profoundly modify psychological research and practice by providing powerful tools for text analysis, conversational assessment, and mental health support for clinicians and patients. Their capacity to interpret and respond to natural language prompts opens up opportunities for psychologists to enhance psychological diagnostics and therapeutic interventions. Patients and the broader public can, and are already doing so, turn to AI chatbots for affordable, on-demand mental health support and therapy.
However, these applications raise ethical, methodological, and legal challenges. Within the EU context, compliance with the General Data Protection Regulation (GDPR) and the EU Artificial Intelligence Act (AI Act) presents pressing concerns. Issues of data minimisation, informed and explicit consent (where applicable), and transparency of algorithmic processing intersect directly with the sensitive nature of psychological data. Additionally, a key question under the AI Act is whether certain uses of LLMs in psychology - whether by psychologists or by patients themselves - might fall under the "high-risk" category, which would trigger more stringent requirements around transparency, oversight, and safety.
Beyond these considerations, the very design of LLMs raises further concerns for psychological practice. Their complexity and opacity, combined with probabilistic outputs, undermine reliability and accountability in clinical use. The risks of bias, reinforcement of stereotypes, and generation of harmful content could have a profoundly negative impact on vulnerable populations. We will critically evaluate the compatibility of LLM usage in psychology with European regulatory requirements. Ultimately, the discussion will provide a framework for integrating LLM technologies into psychological science and practice under the evolving European legal landscape.