Policy briefing - An EU AI Act that works for people and society

An EU AI Act that works for people and society 8 (‘Of the five most populated countries in the EU, only Spain, with a new regulatory AI sandbox and AI regulatory agency, seems to be well prepared’)9 and reduce the risk of divergent levels of protection as seen at times under the GDPR. [...] • Ada supports the European Parliament’s proposal to set up permanent sub-groups of AI developers and other relevant stakeholders in the AI Office Advisory Forum, to consider the governance of foundation models and R&D. [...] These challenges centre on: • the complexity of the AI value chain, and the number of actors building on top of foundation models • the multi-functionality of foundation models (they can be deployed for a variety of uses with divergent risk profiles) and the tendency for capabilities – and therefore risk – to grow as the models are trained with more data and computing resources An EU AI Act that w. [...] Such reporting should be coupled with early or priority access to models for research and safety purposes, as recently offered by leading labs in the UK.25 Mandatory disclosure and early access for safety would reduce information asymmetry and offer regulators prior warning of advancements in ‘state of the art’ capabilities and the ability to better prepare for the impact of these developments.26. [...] We therefore echo leading AI researchers in strongly welcoming the avenues for remedy and redress introduced in Chapter 3a of the European Parliament’s text, such as a right to lodge a complaint with a supervisory authority, judicial remedy and an explanation of individual decision-making.82 These must be included in the final Act, to offer recourse for (harmed) people and also to complement the f.