After several months of internal negotiations, the European Parliament's political groups finally agreed, on Thursday 27 April, on a common position for the future interinstitutional negotiations on the Artificial Intelligence Act (‘AI Act’) (see EUROPE 13167/6). After being postponed several times, the vote in the European Parliament committee is expected to take place on 11 May. The plenary vote could take place in June.
Recent developments and the deployment of new artificial intelligence (AI) systems have prompted the rapporteurs to look at the issue of AI for general use. MEPs felt that more needed to be done on this issue. Some even favoured the option of a new specific legislative proposal (see EUROPE 13168/12).
On this point, MEPs said that general-purpose AI systems, such as the ChatGPT generative AI system, should be subject to additional requirements before they are put on the market, in particular to ensure that they have been developed in accordance with existing EU laws on fundamental rights.
MEPs also returned to the issue of classifying certain AI systems as ‘high risk’. Thus, the text provides that only systems that pose a ‘significant’ risk to health, safety and fundamental rights are considered 'high risk'. They would be subject to stricter transparency and governance rules.
These systems, which are classified as ‘high-risk’, should also be subject to an impact assessment, taking into account the consequences for fundamental rights, and should comply with the relevant European environmental standards and legislation.
In addition, MEPs also agreed on prohibited practices under future AI legislation. Thus, AI systems whose purpose is biometric identification could only be used a posteriori in cases of “serious crime”. The use of such systems should be subject to prior judicial approval.
The use of emotional recognition systems in law enforcement, labour, health and migration management operations would be prohibited. (Original version in French by Thomas Mangin)