The Italian Competition Authority (AGCM) has strengthened consumer protection oversight in relation to generative artificial intelligence services. Through three investigations concerning DeepSeek, Mistral AI and NOVA AI, the Authority addressed the issue of so-called AI “hallucinations”, namely the generation by chatbots of inaccurate, misleading or fabricated content. The proceedings were closed with the acceptance of the commitments proposed by the companies, without any finding of infringement pursuant to Article 27(7) of the Italian Consumer Code.

The commitments undertaken are aimed at making notices to users clearer, more accessible and more timely, both on websites and in mobile applications, including at the stages preceding registration or purchase of the service. In particular, the companies will be required to include permanent disclaimers in the user interfaces, below the chat windows, in Italian and with dedicated information links, warning users that AI-generated responses may contain errors or unreliable information.

The Authority also required the strengthening of pre-contractual information, through explicit warnings concerning the reliability limitations of content produced by AI systems and the need to verify its accuracy. In the case of DeepSeek, the commitments also include a technological investment aimed at mitigating the phenomenon, while acknowledging that, at the current state of the technology, hallucinations cannot be entirely eliminated.

The intervention concerning NOVA AI is also specific in scope: the service will have to provide clearer information to consumers regarding its nature as a single access interface to different chatbots, without presenting itself as an autonomous system for aggregating or reworking responses.

In its decisions, the AGCM emphasised that information concerning the risk of inaccurate responses is relevant to enabling users to make an informed choice, including where the service does not require payment of a monetary consideration.

The matter confirms the extension of unfair commercial practices enforcement to generative AI services and highlights the increasing relevance of transparency obligations in the provision of AI-based consumer-facing tools.