AI Act after August 2026: turn the current timeline into a project plan

August 2026 was an important AI Act milestone, but it is not a single deadline for every AI project. The July 2026 amendment changed key dates. A company's plan should identify the applicable provision and system before assigning a delivery date.

The dates below were checked on 11 October 2026. They are a planning reference for discussion with the person assessing your obligations, rather than a determination that a particular rule applies to your application.

The amended milestones

The European Commission states that the AI Omnibus entered into force on 27 July 2026. Its announcement moves the application of Annex III high-risk rules to 2 December 2027 and those for high-risk AI embedded in Annex I regulated products to 2 August 2028. Commission announcement

The current Service Desk timeline gives the following milestones:

The Article 50(2) transition concerns certain synthetic-content systems already placed on the market before 2 August 2026. It should not be read as a blanket extension for all transparency work. The timeline also records earlier milestones for prohibitions and general-purpose AI models. Official implementation timeline

Two projects should not share an unexplained deadline

Consider a fictional company planning a public support chatbot and a separate application for recruitment. Someone adds both to a spreadsheet under “AI Act deadline”. The sheet does not say which provision applies or what must be delivered.

The project owner needs two descriptions and an assessment for each. The chatbot may prompt a transparency review. The recruitment application's purpose and use need classification before the team schedules the corresponding work. This example does not establish the legal category of either real system.

A practical plan records the source date, the assessed version and the person who confirmed the requirement. If an amendment changes a milestone, the team can identify which tasks need reassessment. Replacing one date at the top of the spreadsheet risks changing unrelated work.

Schedule the evidence as well as the implementation

Some tasks depend on supplier information or a decision by the customer. The plan should show those dependencies. An engineer cannot demonstrate that an agreed requirement is met when nobody has specified which deployment or feature the requirement covers.

For the fictional chatbot, the work might include agreeing the disclosure text, implementing it and checking every entry point. A separate task would record the review and its scope. A later channel integration would need its own acceptance decision instead of inheriting a tick from the website release.

Keep internal target dates distinct from statutory dates. A company may choose an earlier target to leave time for review or procurement. That useful planning choice should not appear in an article or supplier brief as a legal deadline.

How Syntalith can help prepare the plan

Syntalith can scope the technical inventory, implementation tasks and acceptance evidence after the relevant requirements have been assessed. The proposed work can identify missing supplier answers and assign technical dependencies to the appropriate team. Legal interpretation and confirmation of applicability remain separately assigned responsibilities.

For the fictional company's two projects, a first engagement could produce separate implementation scopes with unresolved questions. The quote should state whether it covers preparation only or also delivery and verification. We do not represent a project plan as certification or proof of legal compliance.

Describe the application and the milestone you are planning around. The first conversation can establish what documentation already exists and which decisions are still missing. See service pricing, the Article 50 interface guide, and the classification preparation guide.