Share This Article
The recent developments on the EU AI Act set high expectations that a political agreement on its contents will be reached soon.
On October 17th, 2023, the Spanish presidency of the EU Council of Ministers shared a significant document in preparation for the upcoming political negotiations with the European Parliament and Commission on October 24th – known as trilogues.
The key focus areas include how to address foundation models—large machine-learning models trained on extensive datasets that generate responses based on specific stimuli. Below is a snippet of the issues currently discussed
🤖 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥𝐬 & 𝐆𝐞𝐧𝐞𝐫𝐚𝐥-𝐏𝐮𝐫𝐩𝐨𝐬𝐞 𝐀𝐈:
One critical aspect under discussion is the regulation of AI models without specific purposes. The Spanish presidency introduces a potential definition of foundation models as AI models capable of competently performing a wide range of distinctive tasks.
📈 𝐕𝐞𝐫𝐲 𝐂𝐚𝐩𝐚𝐛𝐥𝐞 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥𝐬:
A new category, ‘very capable foundation models,’ is emerging. These models, with advanced capabilities, may require additional obligations due to their cutting-edge nature.
🖋 𝐂𝐨𝐩𝐲𝐫𝐢𝐠𝐡𝐭 & 𝐃𝐞𝐭𝐞𝐜𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲:
The presidency emphasizes the need for foundation model providers to demonstrate compliance with EU copyright law, including allowing rightsholders to opt out. For generative AI models, Spain wants providers to ensure their output is detectable as artificially generated.
🏛 𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞:
EU member states initially left AI law enforcement largely in the hands of national authorities. However, recognizing the complexity of AI models and systems, especially foundation models, the presidency has accepted the idea of an AI Office, tasked with overseeing rules on foundation models and large-scale general-purpose AI systems, defining auditing procedures and carrying out compliance controls and investigations.
🔍 𝐁𝐢𝐨𝐦𝐞𝐭𝐫𝐢𝐜 𝐈𝐝𝐞𝐧𝐭𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧:
A significant contention arises regarding the use of real-time biometric identification systems by law enforcement. Negotiations continue to narrow exceptions, with an emphasis on safeguards and judicial authorization for system usage.
🚫 𝐁𝐚𝐧𝐧𝐞𝐝 𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐞𝐬:
The EU Parliament’s mandate includes bans on emotion recognition, biometric categorization of protected data, and predictive policing in various contexts.
Given the recent developments, a political agreement on the AI Act will be reached by the end of the year, and this is the intent of the Spanish presidency of the EU. The European Union wants to take a leading role on artificial intelligence and the adoption of clear set of rules is a pivotal milestone.
On the same topic, you can read the article “What cybersecurity obligations under the EU AI Act”.