The EU AI Act Deadline Was Postponed. Your AI Architecture Wasn't.
Europe moved its hardest rules by sixteen months. What that actually tests — and the one rule that did land. On 2 August 2026, the EU AI Act's obligations for high-risk systems were finally due to apply. Risk management. Data governance. Record-keeping. Human oversight. The parts with real engineering consequences, after two years of preparation. Six days before that date, they moved. Regulation…
The EU AI Act deadline for high-risk systems was postponed by sixteen months, with obligations now applying until December 2027 instead of the originally scheduled date of August 2026. This delay was not due to the rules being unreasonable, but rather because the necessary infrastructure for enforcing compliance, such as harmonised technical standards and accredited conformity assessment bodies, did not exist by the original deadline.
However, one important rule did take effect, which requires systems intended to interact directly with people to inform users that they are interacting with an AI system, unless this is obvious. Synthetic audio, image, video, and text must also be marked in a machine-readable format and detectable as artificially generated or manipulated.
Brief written by urgent.news from Dev.to's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
