The Civil Liberties (LIBE) and Internal Market (IMCO) committees of the European Parliament have endorsed a provisional agreement on landmark AI rules ahead of a vote by the legislative assembly in April. The committees voted 71-8, with seven abstentions, in favour of the compromise negotiated with EU member states in trilogue talks held at the end of the previous year.

Earlier this month, 27 EU countries unanimously approved a deal reached in December on the AI Act, which aims to better control governments' use of AI in biometric surveillance and how to regulate AI systems.

On 22 January, a draft of the AI Act was leaked by Luca Bertuzzi, a journalist from Euractiv. In the latest version, the wording has changed; most notably, it now contains an exemption for open-source AI models -systems with freely available source codes and training data. Although these tools operate more transparently than the so-called 'black-box' systems such as OpenAI's GPT-4, experts still warn that they are capable of harm. 

About the EU AI Act

The EU AI Act, initially introduced by the Commission in April 2021, establishes guidelines for AI developers contingent on the capabilities of their models and/or the intended application of AI. The act outlines prohibited uses of AI and sets rules specifically for identified high-risk applications, including domains like education, health, or employment. These rules encompass obligations related to data quality, testing, and risk assessment. Moreover, the act imposes transparency requirements on general-purpose AIs and tools like deepfakes and AI chatbots.

What is next?

The next steps in the process involve the European Parliament's approval at the plenary session, which is anticipated for 10-11 April -to be followed by ministerial approval. Once adopted, the AI Act will be formally signed and become EU law, triggering an implementation period before it takes effect.

The EU AI Act will be fully applicable 24 months after entry into force, except bans on prohibited practises, which will apply six months after the entry into force; codes of practise (nine months after entry into force); general-purpose AI rules, including governance (12 months after entry into force); and obligations for high-risk systems (36 months).