Brussels switched on AI labelling in August and pushed the high risk rules out to December 2027

The European Union's Artificial Intelligence Act became applicable on 2 August 2026. The parts of it that govern hiring, credit scoring, education and biometric identification did not. On the European Commission's own account of the regulatory framework, high risk AI systems "will be subject to strict obligations before they can be put on the market" from 2 December 2027, and systems embedded in regulated products such as lifts and toys from 2 August 2028.

The deferral came through the package the Commission calls the AI Omnibus, part of a wider simplification effort known as the Digital Omnibus. The Commission's page records it as adopted on 19 November 2025, reaching political agreement on 7 May 2026, and entering into force on 27 July 2026. The same package created a prohibition on AI systems that generate non consensual sexually explicit and intimate content, or child sexual abuse material, which the Commission dates to December 2026. That prohibition has not yet taken effect.

What did switch on this month are the transparency duties. Article 50 of the AI Act requires providers to tell people when they are interacting with an AI system unless that is obvious from the context, and to mark synthetic audio, image, video and text in a machine readable format that is "effective, interoperable, robust and reliable as far as this is technically feasible". Deployers must disclose deepfakes, with carve outs for artistic, satirical and fictional work, and must disclose AI generated text published on matters of public interest unless a person has taken editorial responsibility for it.

The commercial compliance trackers that many companies read instead of the statute do not tell the same story. Kiteworks, in a guide marked updated 6 July 2026, states that "The European Parliament approved amendments on June 16, 2026, pushing the compliance deadline for standalone high-risk AI systems under Annex III from August 2, 2026 to December 2, 2027." That lands on the same destination the Commission gives, by a different route and a different date. The same page also states that the ban on AI powered non consensual intimate imagery applications "took effect December 2, 2026", a date that had not arrived when the page was updated and has not arrived now.

Collibra, writing on the same subject, was more cautious and less precise. It describes "a provisional 'Digital Omnibus' agreement reached in spring 2026" that "deferred several high-risk obligations", says use based high risk systems gained time beyond August 2026 and product regulated systems gained time beyond their later deadline, and warns that the deferrals "are not yet final and should be confirmed". On the Commission's account they were confirmed on 27 July 2026, before Collibra's caveat would have been read by anyone planning an August compliance programme.

The penalty headline that travels with these stories is also narrower than it looks. Article 99 does provide for fines up to 35 million euro or 7 per cent of total worldwide annual turnover, whichever is higher, but it reserves that ceiling for prohibited AI practices. Breaches of operator obligations, which is where most providers, importers, distributors and deployers sit, and which expressly includes the transparency obligations that started this month, are capped at 15 million euro or 3 per cent. Supplying incorrect, incomplete or misleading information to a notified body or a competent authority carries up to 7.5 million euro or 1 per cent. Small and medium enterprises are capped at the lower of the euro figure and the percentage rather than the higher.

Some reference material has simply not caught up. The implementation timeline published at artificialintelligenceact.eu still sets out the pre omnibus statute, under which the remainder of the Act applied from 2 August 2026 and Article 6(1) from 2 August 2027, and carries a note that it was last updated on 1 August 2024.

What is not yet known is whether the December 2027 date holds, and who will enforce it. The AI Office was advertising 40 new posts in technical, legal and operations roles dedicated to AI Act enforcement, with expressions of interest closing on 8 September 2026. Neither the Commission page nor the vendor guides reviewed here says what happens to a high risk system placed on the market between now and December 2027.