EU AI Act under pressure
Civil-society groups warned EU lawmakers that proposed simplifications to the AI Act risk weakening protections for high-risk systems, and European Digital Rights published an open letter urging lawmakers to preserve the original safeguards. A consultant piece also argued implementers may respond by moving models on-premise to reduce transfer and provider risks. (dig.watch) (edri.org) (seresa.io)
European civil-society groups are pressing European Union lawmakers not to reopen the Artificial Intelligence Act in ways that would cut rules for high-risk systems. (dig.watch) European Digital Rights, a Brussels-based rights group, published an open letter warning that “simplification” should not remove obligations for systems used in areas such as hiring, education, policing, migration, and access to public services. The group said those uses can affect jobs, benefits, and freedom of movement. (edri.org) The Artificial Intelligence Act is the European Union’s risk-based rulebook for artificial intelligence systems. It bans some uses outright, puts the strictest duties on “high-risk” systems, and sets separate obligations for general-purpose models that can be adapted for many tasks. (artificialintelligenceact.eu) The law entered into force on August 1, 2024, but its requirements arrive in stages rather than all at once. Rules on prohibited practices started applying on February 2, 2025, and rules for general-purpose artificial intelligence models are scheduled to apply from August 2, 2025, with more obligations for high-risk systems following later in the rollout. (artificialintelligenceact.eu) That timetable has turned implementation into the next political fight in Brussels. Rights groups are trying to lock in the original safeguards before the European Commission and national authorities turn broad legal text into codes of practice, standards, and enforcement decisions. (edri.org) Business advisers are making the opposite case on some compliance choices: Seresa, a consulting firm, argued that companies using outside model providers may reduce transfer, confidentiality, and vendor risk by running models on their own infrastructure instead of sending data to third-party clouds. (seresa.io) That does not remove the law’s duties. The Act regulates how a system is used and how risky that use is, so an employer screening applicants or a public body ranking people for services can still face high-risk obligations even if the model runs on-premise. (artificialintelligenceact.eu) The immediate question is not whether Europe will have an Artificial Intelligence Act, but how much of its original architecture survives the implementation phase. Civil-society groups are trying to keep the guardrails in place before the most consequential parts of the law fully bite. (dig.watch)
