EU AI Act pressures grounding
- Times Higher Education reported April 27 that European universities may have to overhaul AI use, after policy officials warned ChatGPT-based grading could breach EU AI Act rules. - The pressure point is classification: student assessment and many hiring tools count as high-risk AI, triggering transparency, documentation, and human-oversight duties for deployers. - Brussels has delayed some high-risk deadlines, but literacy and model rules already apply, keeping compliance pressure on buyers. (digital-strategy.ec.europa.eu)
European universities are being told their everyday use of tools like ChatGPT may not survive the European Union’s AI Act unchanged. (timeshighereducation.com) Thomas Jørgensen of the European University Association told Times Higher Education that academics using ChatGPT to assess student work could face legal risk because the European Commission’s AI Office has not yet issued the guidance institutions need. (timeshighereducation.com) Under the EU AI Act’s risk-based system, AI used for student assessment sits in the same high-risk bucket as AI used in hiring and credit decisions. (timeshighereducation.com) (digital-strategy.ec.europa.eu) High-risk status is the part buyers care about because it brings concrete duties: documentation, transparency, risk controls, and human oversight rather than a silent algorithm making the call alone. (crowell.com) (digital-strategy.ec.europa.eu) That is pushing organizations toward grounded systems that can show where an answer came from, what data informed it, and where a human can stop or approve an action. This is partly a product-design response to legal duties that ask deployers to explain and supervise AI outputs. (crowell.com) (digital-strategy.ec.europa.eu) The timing is awkward because the law is already live in phases. The AI Act entered into force on August 1, 2024, Article 4’s AI-literacy duty started on February 2, 2025, and general-purpose AI model obligations started on August 2, 2025. (digital-strategy.ec.europa.eu) (artificialintelligenceact.eu) (digital-strategy.ec.europa.eu) The biggest 2026 question is whether the strictest high-risk obligations will hit in August or later. Parliament backed a delay in March 2026, moving stand-alone high-risk systems to December 2, 2027 and some product-linked systems to August 2, 2028, but the final law still depends on the rest of the EU process. (europarl.europa.eu) (aiactblog.nl) (addleshawgoddard.com) That possible delay has not removed the pressure. Lawyers advising employers say many human-resources tools are still likely to be classified as high-risk, and companies still need to map systems, train staff, and prepare consultation and oversight processes. (crowell.com) For universities, that means the informal phase of generative AI is ending. If an institution cannot show what the system is doing, what evidence supports an output, and where a person remains accountable, “change everything” is no longer just a warning line. (timeshighereducation.com)
