The European Union’s ambitious efforts to regulate artificial intelligence (AI) face a major stumbling block, as negotiators struggle to reach a consensus on the treatment of foundation models, specifically generative AI systems like ChatGPT. Six sources with knowledge of the ongoing discussions revealed to Reuters that the impasse threatens the landmark legislation designed to keep AI in check.
Scheduled for final talks on December 6, negotiators are set to tackle crucial issues that have emerged during discussions. The focus has shifted significantly to the regulation of ‘foundation models,’ the heart of generative AI systems, with Microsoft-backed OpenAI’s model being a notable example. These AI systems are trained on extensive datasets, possessing the capability to learn and adapt to new information for diverse tasks.
The proposed AI Act, after two years of negotiations, received approval from the European Parliament in June. However, the path forward requires agreement through meetings involving representatives from the European Parliament, the Council, and the European Commission.
Sources, who preferred to remain anonymous due to the confidentiality of the discussions, highlighted that the divisive issues include the regulation of foundation models, access to source codes, imposition of fines, and other related topics. This week, experts from EU member countries are convening to solidify their position on these matters, coinciding with the gathering of European Parliament lawmakers to finalize their stance.
The major point of contention revolves around the regulation of foundation models, particularly the definition of which models should fall under strict regulatory scrutiny. Some experts and lawmakers advocate for a tiered approach, suggesting that models with more than 45 million users should face more stringent regulations. However, opposing voices argue that even smaller models can pose significant risks and should not be overlooked.
The urgency of reaching a consensus is heightened by the looming threat of European parliamentary elections next year. If negotiators fail to agree on key provisions of the AI Act, the entire initiative risks being shelved due to insufficient time for further deliberations.