EU Parliament Passes Tightened AI Act, Imposing New Human Oversight
In Brussels on Thursday, the European Parliament voted 350‑30 to adopt a revised AI Act that introduces mandatory human oversight for high‑risk artificial intelligence systems. The new rules will require developers to conduct bias audits on datasets and provide clear documentation for regulators. This legislation marks the first time the EU has imposed legal duties on AI developers beyond product safety. The decision follows months of debate over the balance between innovation and accountability.
What Happened
On September 10, 2026, the European Parliament convened in Brussels to vote on the revised AI Act. The bill, drafted by the European Commission and amended by the Parliament’s Committee on Industry, Research and Energy, received 350 votes in favor, 30 against, and 10 abstentions. The vote was conducted via electronic ballot after a five‑minute debate, a procedural change adopted to expedite the process. Key provisions include mandatory human oversight for high‑risk AI systems, a comprehensive bias‑audit requirement for training datasets, and a public registry of certified high‑risk AI applications. The act also establishes an EU AI Oversight Authority to monitor compliance and impose sanctions. According to the official statement released by the European Parliament, the law will take effect on 1 January 2027.
The legislative package is the culmination of a two‑year consultation process that involved more than 2,000 stakeholders, including industry leaders such as DeepMind, OpenAI, and SAP, as well as civil‑society groups and consumer advocates. The European Commission’s Director-General for Digital Policy, Maria Lopez, highlighted that the act aims to "
Despite stricter rules, a 2026 EU study found that AI systems with mandatory human oversight actually generated 12% fewer biased outcomes than fully automated models, overturning the assumption that human involvement always slows progress.
