EU AI Act deadlines and compliance calendar

Article 113 is the adopted-law anchor for the AI Act calendar. The Regulation applies generally from 2 August 2026, selected chapters applied earlier, and Article 6(1) with corresponding obligations has a 2 August 2027 baseline in the adopted text. Later legislation can amend particular dates, so retain the baseline until the final amending act is verified.

Use one calendar row per application date. Each row should identify the legal provision, affected systems or models, accountable owner, required evidence, and the internal readiness date that comes before the legal application date.

The first operating deadline is not a paperwork date. It is the point when Chapter I and Chapter II apply, so the calendar should force two workstreams: an AI literacy owner for staff and other persons operating or using AI systems on the organisation's behalf, and a prohibited-practices owner for product and deployment screening.

The useful evidence is concrete: training audience, AI system context, role-specific literacy material, product-screening notes, and a release gate showing that prohibited-practice review happened before placing on the market, putting into service, or use.

The 2 August 2025 row should separate GPAI model-provider obligations from ordinary AI-system deployment work. Article 113 brings Chapter V into application on this date, while Commission GPAI guidance states that GPAI provider obligations enter into application on 2 August 2025.

For organisations building on third-party GPAI models, the calendar should still assign a procurement and product owner. They need model documentation requests, downstream integration records, supplier contact points, and evidence that the product team knows whether the organisation is only deploying a model, integrating it into an AI system, or significantly modifying it.

Use 2 August 2026 for the adopted Regulation's general application date, Article 50 transparency readiness, the start of Commission enforcement powers for GPAI provider obligations, and the Article 57 regulatory-sandbox deadline. These are different workstreams even though they share a date.

Keep the adopted 2 August 2027 Article 6(1) baseline and the 2 August 2027 transition for GPAI models already on the market before 2 August 2025. Separately record Commission guidance following the AI Omnibus political agreement: 2 December 2027 for specified high-risk areas and 2 August 2028 for product-integrated high-risk systems. Do not replace the enacted-law baseline until the final amending text is formally adopted and verified.

Article 111 is not a general grace period for every AI system. It creates specific transitional treatment for large-scale IT systems, pre-2 August 2026 high-risk systems, public-authority high-risk systems, and GPAI models placed on the market before 2 August 2025.

Keep these rows separate from Article 113 because the owner question is different: not just when a chapter applies, but whether an already-placed system or model is grandfathered, changed, public-authority-facing, or subject to a later compliance deadline.

A useful AI Act calendar is more than a date list. Each row should be tied to a system or model inventory view so product and compliance teams can see which obligations are approaching, which evidence is missing, and who can unblock the work.

Do not validate unsupported future guidance dates in the calendar. If a date is not in the cited source material, leave it out and track the dependency as a watch item rather than a legal milestone.