Central Development
On 7 May, the European Parliament and the Council reached a provisional agreement to streamline parts of the EU’s AI Act and add targeted prohibitions. The package narrows the definition of a “safety component,” removes overlapping requirements for certain machinery products, and bans AI systems that create child sexual abuse material as well as so‑called “nudifier” apps, according to the European Parliament. The deal also permits processing of personal data to detect and correct AI bias under safeguards, delays watermarking obligations for AI‑generated content until December 2026, and staggers new duties for high‑risk systems to December 2027 and August 2028. The aim is to reduce regulatory friction while maintaining core protections, the European Commission said.
Why It Matters
The agreement signals a recalibration toward enforceable risk controls paired with clearer, longer implementation timelines. Narrowing what counts as a safety component and removing overlaps could lower compliance costs for manufacturers, while explicit bans on CSAM‑creating and nudification tools tighten guardrails where harms are well‑evidenced. Allowing limited personal‑data use for bias mitigation addresses a persistent tension between accuracy and privacy. Phased obligations and delayed watermarking give industry more lead time to adapt systems and governance.
Perspective
Official messaging emphasizes simplification to “boost innovation” alongside citizen protection, per the European Commission. By contrast, some coverage framed the compromise as watering down rules, reflecting concern that extended deadlines dilute urgency; the aggregator Ground News highlighted that characterization. The core legal text and annexes—once published—will clarify how much regulatory scope has actually shifted versus being sequenced over time.
What to Watch
Publication of the final legal text and formal adoption steps in Parliament and Council.
- Commission and regulator guidance on permitted personal‑data use for bias mitigation.
- Implementation calendars for watermarking and high‑risk system obligations (Dec 2026; Dec 2027/Aug 2028).
- National enforcement coordination on bans targeting CSAM‑creating systems and nudification apps.
