Back to News
    Regulatory update

    The AI Act now has fixed dates — what Regulation (EU) 2026/1744 actually changed.

    On 24 July 2026 the AI Act stopped depending on a Commission decision on standards readiness. The enacted amending regulation replaces that trigger with fixed calendar dates: 2 December 2027 for Annex III high-risk systems and 2 August 2028 for product-embedded high-risk systems under Annex I. Here is what changed, who it reaches, and what to do this quarter.

    What actually changed in July.

    Regulation (EU) 2026/1744 was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It amends the application timetable of the AI Act. The change is not a delay and not a softening — it is a change of mechanism.

    The proposal (COM(2025) 868) tied the high-risk obligations to a Commission decision on standards readiness, plus six months. That construct was removed in trilogue. The enacted text sets unconditional calendar dates.

    For anyone planning a compliance programme this is the single most important detail: there is no longer a condition to wait for. The dates are fixed.

    The dates as enacted.

    • 2 February 2025 — prohibited practices apply. Already in force.
    • 2 August 2025 — obligations for general-purpose AI models apply. Already in force.
    • 2 August 2026 — transparency obligations apply.
    • 2 December 2026 — the new prohibitions introduced by the amending regulation apply.
    • 2 December 2027 — high-risk obligations for Annex III systems apply.
    • 2 August 2028 — high-risk obligations for product-embedded systems under Annex I apply.
    • 2 August 2030 — legacy systems operated by public authorities must be brought into conformity.

    Models placed on the market as general-purpose AI are subject to systemic-risk obligations where training compute exceeds 10^25 floating-point operations. Maximum sanctions remain up to 7 percent of total worldwide annual turnover or EUR 35 million, whichever is higher.

    Why the mechanism matters more than the date.

    A trigger-plus-six-months structure cannot be planned against. It can only be watched. Every internal roadmap built on it inherits an unknown start date, which in practice means the work does not start.

    An unconditional date does the opposite. From 27 July 2026 a supplier can count backwards from 2 December 2027 and know exactly how much time is left for classification, conformity assessment, technical documentation and post-market monitoring.

    This is also why an assessment produced before 24 July 2026 can be internally consistent and still wrong. It was correct against the proposal. It is not correct against the enacted text.

    Who this reaches through the value chain.

    The AI Act does not only reach model developers. It reaches providers who place a system on the market under their own name, deployers who use a high-risk system in a professional capacity, importers, distributors, and manufacturers who embed an AI system in a product already covered by Union harmonisation legislation under Annex I.

    • A machinery or medical device manufacturer embedding an AI safety component falls under the Annex I route, with the 2 August 2028 date.
    • An industrial supplier deploying an AI system for recruitment, worker management or credit decisions falls under Annex III, with the 2 December 2027 date.
    • A company that rebrands a third-party model under its own name becomes a provider, with the full provider obligations.
    • A company that only uses AI internally for non-listed purposes is largely reached by the transparency limb from 2 August 2026, not by the high-risk regime.

    The distinction between these roles decides which obligations apply and from when. It is a placement question, not a technology question.

    How Regweaver helps.

    Regweaver is built as a three-step ladder on one shared foundation. RegCheck establishes which of 45+ frameworks reach a specific company and from what date. RegWatch monitors the underlying legal acts daily against the EU source. The platform turns the result into operational control across the value chain.

    The AI Act amendment is a direct example of what that monitoring is for. Regulation (EU) 2026/1744 changed a mechanism, not a topic. A keyword alert would have reported it as news; a source-monitored register reports it as a change to the dates a specific assessment depends on.

    Every date in this article is carried in the underlying data with a verbatim quote from the consolidated legal text and a reference to the act it comes from. Quality-reviewed against consolidated legal text — not summarised from secondary reporting.

    What to do this quarter.

    • Date every existing AI Act assessment. Anything produced before 24 July 2026 was written against the proposal and needs to be re-run.
    • Decide your role per system: provider, deployer, importer, distributor, or manufacturer under Annex I. The role, not the model, determines the obligations.
    • Separate the Annex III inventory from the Annex I inventory. They have different dates — 2 December 2027 and 2 August 2028 — and different conformity routes.
    • Prepare the transparency limb now. 2 August 2026 is the nearest binding date and it reaches far more companies than the high-risk regime.
    • Put the amending regulation under monitoring rather than in a slide. Implementing and delegated acts under the AI Act will continue to arrive.

    Key takeaway

    A date you can plan against is worth more than a condition you can only watch. From 27 July 2026 the AI Act high-risk timetable is unconditional — and every assessment written against the proposal needs to be re-run.

    Monitor EU-AI-ACT

    We watch for changes and send a reviewed notice when something actually changes.

    See the monitoring option

    This article is for informational purposes only and does not constitute legal advice.

    Related reading