Legal teams read the EU AI Act as a set of obligations and fines. Engineers read it as data schemas, evaluation suites, and log retention. Bridging the two is the most practical step most organisations can take before the next set of dates.

The 2025–2028 Timeline

The Act entered into force on 1 August 2024. The European Commission AI Act Service Desk timeline, which reflects the Digital Omnibus on AI, sets out the staged application:

  • 2 February 2025: General provisions, including AI literacy measures (Article 4), and the prohibitions in Article 5 apply.
  • 2 August 2025: Rules for general-purpose AI models apply. The penalty provisions (Chapter XII) also apply from this date, and Member States must have their penalty rules in place.
  • 2 August 2026: Article 50 transparency obligations apply. The Commission has also said that enforcement starts for several rules, including prohibitions, transparency, and AI literacy.
  • 2 December 2026: New prohibitions apply, including for non-consensual sexual deepfakes and child sexual abuse material. The Article 50(2) transition also ends for providers whose generative AI systems were placed on the market before 2 August 2026.
  • 2 December 2027 and 2 August 2028: Obligations for standalone Annex III high-risk systems apply from 2 December 2027, and for high-risk AI embedded in Annex I products from 2 August 2028.

Article 50 and the 2 December 2026 Transition

Under Article 50(2), providers of AI systems that generate synthetic audio, image, video, or text must ensure that outputs are marked in a machine-readable format and can be detected as artificially generated. The transition in Article 111(4) is narrow. It covers providers whose systems were placed on the market before 2 August 2026, who must comply by 2 December 2026. A system placed on the market after 2 August 2026 must be marked from the start.

Providers must design systems that interact with people so that users are informed they are interacting with AI (Article 50(1)). Deployers have separate duties under Article 50(4): to disclose deepfakes and AI-generated text published to inform the public.

Engineering Work for Annex III Systems

If your software ranks job candidates, scores creditworthiness, prices life or health insurance, or decides access to education, the Annex III rules apply from 2 December 2027. Engineering work includes:

  • Data governance (Article 10): bias examination, representativeness checks, and lineage across training, validation, and test datasets.
  • Human oversight (Article 14): interfaces that let a competent person understand the output, decide not to rely on it, and intervene, including through a stop control or a similar procedure.
  • Accuracy, robustness, and cybersecurity (Article 15): adversarial testing against prompt injection, data poisoning, and drift.

Building Technical Documentation Into Releases

Annex IV technical documentation is easier to keep current when parts of it are generated from the build. We often set up release steps that collect model and data versions, evaluation results, and risk-register changes. A person still reviews and signs off the content; the pipeline keeps the evidence from drifting out of date.

Last reviewed 17 November 2026*