Obligation tracker
The tracker takes a role, a classification tier, whether the organisation is established in the Union and, for a deployer, whether Art. 27(1) applies. The tracker returns one register: every obligation the matching profiles carry, its documentation and process elements, the date from which each applies, rendered from the regulation data In force, and a status column. The tool runs in the browser; nothing entered is sent. The register downloads as CSV, a working document for the organisation's own file. Regulation (EU) 2026/1744 moved dates and left obligations unchanged.
How the result is built
The form asks four questions from tracker.json: the role (provider, including a product manufacturer under Art. 25(3), deployer, importer, distributor, or general-purpose AI model provider), the classification tier (seven options, from high-risk under Annex III to a general-purpose AI model with systemic risk), whether the organisation is established in the Union and, for a deployer only, whether it falls within Art. 27(1). The engine selects the profiles in profiles.json whose role and tier match, adds the profile marked for every role (Art. 4, AI literacy), and adds the flag-driven profiles: Art. 22 (authorised representative) for a provider established outside the Union with a high-risk system, Art. 54 for a general-purpose AI model provider established outside the Union, Art. 27 for a deployer within Art. 27(1) of an Annex III system, and Art. 51, Art. 52 and Art. 55 in addition to Art. 53 for a model with systemic risk. The role, tier, establishment and Art. 27(1) answers can also arrive in the page address; the classifier opens the tracker that way with its result filled in.
Each profile lists articles with the regulation entry that dates them. An article with an obligation page expands from data/obligations.json into one row for the article's restated requirement, then one row per documentation element and one per process element, marked in the Kind column. Testable elements are left to the checklist generator, which lists them with the protocol that covers each. An article without an obligation page, such as Art. 16, Art. 43, Art. 49 or Art. 72, carries the label from the profile only. The Applies from column renders the date of the entry named on the row, and the Source column renders the verified-on stamp with the source link, so an Annex III provider sees the Annex III application date Deferred on Art. 9 to Art. 17 and the AI literacy date In force on Art. 4. An Annex I manufacturer sees the Annex I application date Deferred on the same articles.
The Status column holds four values from tracker.json: Not started, In progress, Evidence ready, and Recorded as not applicable. The value and the system name live in the browser only and leave it in the CSV the visitor downloads. The CSV has eight columns: system, article, element, kind, applies-from date, entry status, source URL and status; the date, the entry status and the URL are read from the regulation entry. Reset clears the answers, the statuses and the system name. The methodology page lists the logic files with their versions next to the protocols.
What the result is not
The register is a working document for the provider's or deployer's own file. It is not a classification decision under Art. 6: the tier is an input the visitor chooses, and the provider's own assessment under Art. 6(4) and its documentation stay with the provider. It is not a conformity assessment under Art. 43, and it is not legal advice; the disclaimer applies to every row. A status of Evidence ready records the visitor's own judgement about a document or a process and is not a test result. Where a row's entry is an unsettled point, the badge carries the label Working reading, and the regulation page shows both readings with their sources.
Where the output goes next
The same role and tier open the checklist generator, which lists the testable elements with the protocol that covers each, and the traceability matrix template, which maps every element to an artifact and an evidence location. The templates supply the outlines for the documents the register names, the Annex IV outline first. The classification and readiness assessment service produces the same register under PR-CLS v1.0, with the role and tier recorded per system from the protocol steps, within the limits on what we cannot do.
Questions
What is the timeline for the EU AI Act?
Art. 113 of Regulation (EU) 2024/1689 sets staged application: Chapters I and II first, then the general-purpose AI obligations in Chapter V, then the general application date that carries Art. 50, and later dates for high-risk systems under Annex III and under Annex I. Regulation (EU) 2026/1744 amended Art. 113(c) to move the two high-risk dates and left the other stages in place. Every date on this site renders from the regulation data with its source, and the tracker shows the date on each row.
What are the EU AI Act deadlines?
A deadline in the Act is the date from which an obligation applies, set by Art. 113 per chapter and tier, together with the transitional dates in Art. 111 for systems already on the market. The deadline for an organisation therefore depends on its role and tier: an Annex III provider, an Annex I manufacturer, a general-purpose AI model provider and an Art. 50 provider each carry a different regulation entry. The tracker renders the entry on every row from the regulation data, with its verified-on stamp and its source.
What is the EU AI Act implementation timeline?
An implementation timeline orders the work that precedes each application date. For a provider of a high-risk system the order follows Art. 16: risk management, data governance, technical documentation, logging, instructions for use, human oversight, accuracy and robustness testing, the quality management system, then conformity assessment under Art. 43 and registration under Art. 49 before placing on the market. The tracker gives the register with a status per element; the sequence and the effort stay with the provider, and the regulation page shows which dates the Omnibus moved.
Every inquiry receives a reply within 24 hours. Scope and availability are discussed per request.