Webdesk: As they prepare for legislative talks next month, the EU parliament is striving to address generative AI and get an agreement on the bloc’s first law by year’s end.
“This is the last thing still standing in the negotiation,” EU Parliament co-rapporteur Dragos Tudorache told Tech Crunch. We’re finishing up now. I hope we close next week and vote in May.”
Google and Microsoft lobbyists have been lobbying for generative AI regulation for months.
Tudorache said MEPs recommend creating regulations in layers to cover duties across the AI value chain, foundational model guardrails, and generative model content concerns.
“It needs to explain how the model was trained,” Todorache said. Data accuracy from biases [etc.].
He explains that foundational models’ second layer must do particular things using their power, versatility, and training. Transparency, training, and testing before market entry are key. “What is their level of diligence and responsibility as model developers?” he asks.
“We’re not inventing a new regime for copyright because there is already copyright law out there,” Todorache says of the third layer. The developer’s model training material must be documented and transparent. So that thereafter, the rights holders might say, “Hey, hold on, what you used my data, you used my songs, you used my scientific article—well, thank you very much that was protected by law, therefore, you owe me something—or no.” We’ll use copyright laws. The AI Act does not replace that. That’s inside.”
The EU AI rulebook is still in development, yet AI is advancing quickly. The Commission’s initial draught advocated risk-banding AI.
A few high-risk use cases will be illegal, but most low-risk apps will not. Middle and third have tolerable safety risks.
