EU AI Act — what’s changing
The law is moving. We track it so you don’t have to.
We track EU AI Act guidance as it evolves — so your readiness reflects the law as it stands today, not last year.
Tracking 6 developments
- 2026-12-01ProposedStandard
Harmonised standards still pending
CEN/CENELEC (JTC 21) standards give a presumption of conformity for high-risk systems. First publications are delayed — expected around Q4 2026.
Source: CEN-CENELEC · JTC 21 Artificial IntelligenceWhat it means: You can’t yet claim conformity ‘by standard’. We’ll flag the moment the first harmonised standards land.
- 2026-08-02ScheduledEnforcement
Enforcement powers begin
From 2 August 2026, enforcement powers apply — including the Commission’s power to fine GPAI providers. Around 24 of 27 Member States have now designated national competent authorities.
Source: European Commission · AI Act regulatory frameworkWhat it means: Penalties become real from this point. A sealed, dated readiness record is your evidence that you acted in good faith.
- 2026-05-07ProvisionalAmendment
Digital Omnibus defers high-risk deadlines
The first proposed amendment to the AI Act. Under the provisional agreement (not yet adopted): high-risk Annex III duties move from 2 Aug 2026 to 2 Dec 2027; Annex I products from 2 Aug 2027 to 2 Aug 2028; the synthetic-content marking grace extends to 2 Dec 2026; two further prohibited practices are added.
Source: European Commission · Digital OmnibusWhat it means: If you’re high-risk you likely have more time — but this is not law yet. We carry these dates as provisional and update them the moment they’re published in the Official Journal.
- 2025-07-24PublishedTooling
Official training-data summary template
The Commission published the template for the Art. 53(1)(d) public summary of the content used to train a general-purpose AI model.
Source: European Commission · GPAI training-content templateWhat it means: Our Training-Data Summary document follows this template, so what you produce matches what regulators expect.
- 2025-07-18PublishedGuidance
Commission guidelines on GPAI scope
Official guidance, in all EU languages, on whether the general-purpose AI obligations apply to a given model.
Source: European Commission · GPAI scope guidelinesWhat it means: Helps you decide whether you’re a GPAI provider at all — our checker applies the same logic.
- 2025-07-10PublishedGuidance
GPAI Code of Practice published
A voluntary tool to demonstrate compliance with the general-purpose AI obligations. Three chapters: Transparency and Copyright (all GPAI), and Safety & Security (systemic-risk models only).
Source: European Commission · GPAI Code of PracticeWhat it means: If you provide a GPAI model, signing the Code is a recognised way to show good faith on transparency, copyright and — for large models — safety.
Tawsik guides you toward compliance. It isn't legal advice.