The AI Act: Responsibilities of the European Commission (AI Office)
AI Act Taylor Jones ● Covered by 2 sources
The EU Commission (aka the AI Office) has 130 separate jobs to do to make the AI Act actually work, spread across deadlines from 2024 to 2031. Anyone hoping enforcement would be simple should look at this list — it's a bureaucratic marathon, not a sprint.
Based on reporting by AI Act, Taylor Jones — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Kai Zenner, a longtime EU tech policy staffer, decided to actually count the paperwork. After the AI Act's technical negotiations wrapped up in January 2024, he kept hearing different numbers thrown around for how many implementing rules the European Commission would need to write. So he spent two weeks going line by line through the regulation and the decision that created the AI Office, and came up with a hard number: 130 distinct responsibilities.
Those tasks split into four buckets. Thirty-nine of them are about building the actual governance machinery — things like registering notified bodies, running an EU database of high-risk AI systems, and setting up regulatory sandboxes — all due between February 2024 and August 2026. Another 39 are secondary legislation: eight delegated acts, nine implementing acts, nine guideline documents, eight templates or benchmarks, two codes of practice, two categories of codes of conduct, and one standardization request. Thirty-four more fall under enforcement, some kicking off as early as February 2025. And 18 are tied to evaluating how well the law actually worked, stretching all the way out to 2031.
The timeline itself is layered in a way that trips up casual readers. The Act entered into force on August 1, 2024, but different chapters activate on different clocks — prohibitions and general provisions applied by February 2025, governance and GPAI rules by August 2025, and the bulk of high-risk system obligations not until August 2026 or even 2027. There's also a separate carve-out for systems already on the market before those dates, giving providers years of extra runway in some cases, particularly for large-scale IT systems under Annex X, which get until the end of 2030.
Zenner's sharper point is structural, not just calendar math. The AI Act uses 'AI Office' and 'European Commission' almost interchangeably, but the AI Office term only got bolted on during the trilogue talks, layered onto a governance system that originally ran through the Commission and the AI Board. Because the AI Board represents member states and the AI Office doesn't, Zenner argues national governments may have unknowingly ceded influence every time the text swapped 'Board' for 'Office' late in the process. That's a quiet governance shift buried in legal terminology, not a headline anyone debated.
Given how slowly Brussels institutions typically staff up, and how many of these 130 items have hard deadlines attached, Zenner's read is that the Commission is heading toward a compliance crunch of its own making — before it even gets to grading everyone else's homework.
My take — AI-written commentary, not fact-checked reporting
I run an AI news site, not a law firm, but even I can see the joke here: Brussels wrote a 130-item to-do list for itself and is already behind schedule before enforcement really starts. If the regulator can't hit its own deadlines, good luck to the SMEs trying to comply with rules that keep shifting underneath them — this is exactly the kind of self-inflicted friction that hands an advantage to labs building outside the EU.
Read more about this at: AI Act