The Member States discussed a version of the French EU Council Presidency’s partial compromise text on general-purpose artificial intelligence (AI) systems (see EUROPE 12950/7) at a meeting of the EU Council’s Telecommunications Working Party on Tuesday 17 May.
Firstly, the text introduces a new definition for such systems. Thus, systems put on the market, including in the form of open source software, designed to perform tasks such as image and speech recognition, audio and video generation, pattern detection, question answering or translation would be considered ‘general purpose’.
The compromise document also provides for these AI systems to be subject to certain obligations, which are less onerous than for systems classified as ‘high risk’. In particular, suppliers would be obliged to follow the conformity assessment procedure for their system and would be required to keep the technical documentation available to the competent national authorities for a period ending 10 years after the system was placed on the market.
In addition, the text also provides that system providers, in the event that a general purpose system could be used by another provider as a high-risk AI system or as a component of a high-risk AI system, shall cooperate and provide the necessary information to enable the latter to comply with their obligations.
However, the compromise text introduces several exceptions to these requirements. Thus, these would not apply if the supplier has “explicitly excluded any high-risk use in the instructions for use or information accompanying the general purpose AI system”.
This exclusion must be made in “good faith” and cannot be “justified if the provider has sufficient reason to believe that the system may be abused”, the document concludes.
See the document (in French): https://aeur.eu/f/1px (Original version in French by Thomas Mangin)