European Commission

AI: the transparency requirement does not apply to developers

Guidance on Article 50 of the Artificial Intelligence Act has been published. The expertise gained in this instance helps to avoid the risk of misunderstanding

Supatman - stock.adobe.com

3' min read

Translated by AI
Versione italiana

3' min read

Translated by AI
Versione italiana

The European Commission is attempting to resolve certain interpretative issues relating to EU Regulation 2024/1689 on artificial intelligence, by publishing, in the Official Journal of the EU, Series C, on 12 November, guidelines on the implementation of transparency obligations for certain AI systems.

Article 50

Article 50 of the EU Artificial Intelligence Act, in line with the risk-based approach, requires deployers and providers of AI systems intended to interact directly with natural persons to design them in such a way that ‘the natural persons concerned are informed that they are interacting with an AI system’.

Loading...

However, in order to assist the competent authorities and interested parties, the Commission has, following consultation with stakeholders, issued practical guidance to ensure that transparency obligations are applied in a ‘consistent, effective, proportionate and uniform’ manner.

Article 50 sets out four transparency obligations applicable to various artificial intelligence systems or their outputs; accordingly, the guidelines are structured in such a way as to examine and flesh out these four obligations, which – as the Commission clarifies – may be applied cumulatively.

Interactions between AI and individuals

Thus, if an AI system generatesimages whilst interacting directly with a natural person, the provider may be required to comply with the obligations set out in paragraphs 1 and 2 of Article 50, which require providers both to develop and design the system in such a way that the natural person is informed that they are interacting with an AI system, and to ensure that the outputs of AI systems ‘are labelled in a machine-readable format and identifiable as artificially generated’. In the interests of personal autonomy and the safeguarding of European values, human dignity, fundamental rights, consumer protection and the credibility of media content.

When the transparency requirement is waived

The guidelines set out the circumstances in which the transparency requirement laid down in Article 50 does not apply. These are situations in which AI-based chatbots are designed to assist with writing and reviewing code solely for professional developers who possess the necessary expertise to avoid misunderstandings when interacting with a human being. Other exclusions include interactive AI systems used exclusively by healthcare professionals, as well as those integrated into household appliances with limited functionality, are excluded.

With regard to the exception for AI systems authorised by law for the purposes of crime-fighting, it is specified that such systems are not exempt from transparency obligations when it comes to police chatbots based on artificial intelligence installed on official police websites or mobile applications that enable people to report offences or lodge complaints; in the case of AI-based hotlines operated by the police or other law enforcement authorities; AI-assisted hotlines for reporting fraud, or AI-based digital portals managed by financial institutions or public authorities where users can report alleged financial offences. The scope of the transparency obligations also covers systems that use virtual assistants to collect witness statements, where victims or witnesses can submit ‘information via an AI-based interface’.

Data protection

We will now turn to clarifications regarding labelling, particularly with regard to obligations relating to data protection. As for the obligations applicable to deployers of generative AI systems in cases where deep fakes and texts generated or manipulated by AI and published to inform the public on matters of public interest, the Commission has provided examples of deep fakes in accordance with Article 3 of the AI Regulation and specified the cases in which an image cannot be defined as a deep fake.

Copyright reserved ©
Loading...

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti