The German government’s announcement of its support for the compromise reached between Parliament and Council negotiators on the AI Act should pave the way for its adoption by the Council of the EU. The European Parliament is expected to vote in mid-February.
Our team in Brussels went through the text of the final agreement and shared with us its first conclusions.
- An extraordinarily complex text:
The strong political pressure to reach an agreement before the European elections has resulted in a text of unprecedented complexity, bringing together the numerous proposals made by the Parliament and the member states in an attempt to reach a compromise. Many provisions seem to have been drafted at the last minute, with many inconsistencies.
- A new regulatory regime for general-purpose AI models:
All providers of general-purpose AI models would have to comply with documentation and information-sharing obligations with users. Providers of the largest or “systemic” models would also have to comply with stricter model evaluation and transparency obligations, an approach which reminds the Digital Markets Act or the Digital Services Act. Some transparency obligations on training datasets could finally apply to all general-purpose AI models.
- A central role for the Commission in the AI Act implementation:
Many of the text’s provisions empower the European Commission to specify crucial aspects, such as adding or removing of use cases from the list of high-risk use cases, by means of implementing acts. The Commission will also be responsible for implementing the rules on general-purpose AI models and monitoring their application. The Commission did not wait for the final adoption of the AI Act to set up the AI Office, a new internal structure provided for by the text to support it in its role as regulator.
Will the AI Act be a new RGPD? It certainly looks like it.
Our Brussels team stands ready to answer any question you would have about the text.
