The Commission reviewed its AI list. Therapy chatbots are the problem it couldn’t classify.
Self-help therapy chatbots are flagged for monitoring.
The European Commission published COM(2026) 234 final on 20 May 2026, its first mandatory review of Article 5 and Annex III under Article 112(1) of the AI Act. The headline finding is that no amendments are needed yet. But the most revealing section is a three-paragraph treatment of self-help therapy AI chatbots: a category of product that has caused documented deaths in Europe, attracted the largest GDPR enforcement action against an AI company in EU history, and still falls entirely outside the high-risk classification regime.
The Commission’s conclusion: monitor closely, collect more evidence, no amendment yet.
This article explains what the Commission found, why the gap exists in the first place, what the harm record looks like, and what the Commission should be watching.
