For two years, 2 August 2026 was circled in compliance calendars across Europe. It was the date on which the core obligations of the EU’s Artificial Intelligence Act — including the rules governing high-risk systems — were meant to switch on across the single market.

The date arrived. Much of what was supposed to come with it did not.

What actually switched on

From 2 August the European Commission’s AI Office, working with national authorities, began enforcing the Act. The transparency obligations took effect on schedule, and they are the part of the law most people will actually notice.

Systems that interact directly with people must make clear that the user is dealing with a machine, unless that is obvious from the context. Deepfakes — images, audio or video generated or manipulated by AI — must be disclosed as artificially generated. AI-generated or altered content must also carry machine-readable markings so that it can be detected automatically rather than only by eye.

The Commission’s stated reasoning is that it is becoming genuinely difficult to distinguish generated content from authentic content, and that this creates risks of misinformation at scale, fraud, impersonation and consumer deception. The obligations are aimed at letting people recognise what they are looking at.

What slipped

The heavier part of the regime — the obligations attaching to high-risk AI systems — has been pushed back by more than a year.

That category is where the Act does most of its work on ordinary life. High-risk systems are those used in decisions about employment, education, access to essential services, credit, law enforcement and migration: the settings where an automated judgement can materially change what happens to a person. The delayed obligations include risk management, data governance, technical documentation, human oversight and conformity assessment.

How the change was made, and how late

The AI Act itself is Regulation (EU) 2024/1689, which entered into force on 1 August 2024 with a deliberately staggered timetable. The change came through a separate amendment package known as the Digital Omnibus on Artificial Intelligence.

The European Parliament adopted the text on 16 June 2026. The Council gave final approval on 29 June. The act was signed on 8 July, published in the Official Journal as Regulation (EU) 2026/1744, and entered into force on 27 July — six days before the deadline it was rewriting.

That timing was tight by design and by necessity. Had publication slipped into August, the letter of the law would still have pointed at 2 August for the high-risk obligations, creating a window in which organisations that had stood down their compliance programmes were technically exposed. It landed in time, but only just.

The argument about it

Supporters of the postponement, including much of the technology industry and a significant bloc within the Commission’s own simplification agenda, argue that the deadline was never realistic. The harmonised technical standards that companies need in order to demonstrate conformity were not finished, and requiring compliance with rules whose measurement criteria do not yet exist produces box-ticking rather than safety. They also argue that Europe risks regulating itself out of a market it does not lead.

Critics — including civil society organisations and a number of MEPs who voted against — argue that the standards were late because the process was under-resourced, not because the obligations were unreasonable, and that a delay obtained through concerted lobbying sets a precedent for the next deadline. Their sharper point is about who bears the cost of waiting: the people subject to automated decisions in hiring, benefits and policing now go more than another year without the protections the Act promised them in 2024.

A third group makes a narrower complaint: organisations that spent two years and considerable money preparing for August 2026 have been penalised relative to those that did nothing.

What it means in practice

Brussels remains the most prominent AI regulator in the world, and the AI Act is still the first comprehensive law of its kind. Its rules on general-purpose models are now applicable, which is why some of the most advanced systems tend to launch in the EU a few weeks after other markets while firms finish their compliance work.

The practical instruction for anyone reading older guidance is blunt: material written before July 2026 that describes 2 August as the day the high-risk regime begins is now wrong. The transparency rules are live and enforceable today. The rest is a date in the future, and one that has already moved once.

Sources

European Commission (digital-strategy.ec.europa.eu and commission.europa.eu); Regulation (EU) 2024/1689; Regulation (EU) 2026/1744; Euronews; Technology.org.