The European Commissioner for Values and Transparency, Věra Jourová, announced, on Monday 5 June, that she wanted to introduce “immediately” a label to enable users to easily and quickly identify content generated by artificial intelligence (AI).
“Systems like ‘ChatGPT’ - or others - are capable of creating images, videos or sounds that look very authentic and are very convincing. These are all new challenges for the fight against disinformation”, declared Ms Jourová as she prepared to meet the signatories of the Code of Practice on Disinformation (see EUROPE 13190/35) (see other news).
“Signatories with services that are likely to disseminate AI-generated disinformation should implement technology to recognise this content and make it clear to users”, she added.
The label sought by the European Commission would be part of the Code of Practice on Disinformation. This provision would then be replaced by the rules of the Digital Services Act (DSA) (see EUROPE 13185/5), which stipulates that “manipulated or generated” content must be recognisable by means of a specific and “visible” mark.
The DSA will come into force on 25 August. The very largest online platforms and search engines - with more than 45 million monthly users in the EU - will therefore have to comply with the new rules.
In addition, the European Commissioner detailed, the signatories of the Code, which is based on a voluntary basis and brings together some forty players in the sector, should “incorporate the necessary guarantees to ensure that these services cannot be used by malicious actors to generate disinformation”.
In addition to the future label, the European Commission is considering other AI strategies. With the European Parliament due to formally adopt its position on 14 June ahead of the forthcoming inter-institutional negotiations, the Commission intends to set up an ‘AI pact’ by the end of the year.
Described by Thierry Breton as “the antechamber to the AI Act”, it would “allow voluntary companies to start implementing rules on artificial intelligence until the regulation comes into force”. In this respect, the Commission hopes that an agreement will be reached between the co-legislators before the end of the year. It would then come into force “at the end of 2025 at the earliest”, said Mr Breton.
“We’re not going to force anyone, just encourage them. The pact would make it possible to assess the risks and the way in which platforms respond to them. We are going to pay attention to data quality, traceability, technical documentation and human supervision”, he detailed. (Original version in French by Thomas Mangin)