The state of AI regulation — 2026 overview

This is the starting point: one overview of where AI regulation stands in 2026, with the detail per topic from here. It is a living document — the dates and the state of play are updated whenever something changes.

The core: one regulation, phased

The AI Act (Regulation (EU) 2024/1689) does not enter into force all at once, but in stages. The formal dates from the regulation itself, alongside the shifts from the Digital Omnibus agreement of 7 May 2026 (until publication in the Official Journal, the formal dates stand):

Component	Formal date	After Digital Omnibus
Prohibited practices (Art. 5) + AI literacy (Art. 4)	2 February 2025	unchanged
GPAI regime, governance and penalties	2 August 2025	unchanged
Transparency (Art. 50)	2 August 2026	2 December 2026
High risk, Annex III	2 August 2026	2 December 2027
High risk, Annex I (products)	2 August 2027	2 August 2028

Put these deadlines in your own calendar: [download the calendar](ai-act-deadlines.ics) or [subscribe to it](webcal://yrproject.nl/en/ai-act-deadlines.ics) — the dates then update automatically whenever something changes.

The five regimes

The regulation works with risk categories. Four carry obligations; the fifth — minimal risk — does not.

• Prohibited practices — eight categories of AI

use that have been banned outright since 2 February 2025, with the highest fines.

• High risk — the heaviest set: risk

management, data governance, documentation, human oversight and conformity assessment.

• GPAI models — a dedicated regime for providers of

general-purpose models, with extra duties for systemic risk (training above 10^25 FLOP).

• Transparency — the duty to make AI recognisable:

chatbots, machine-readable marking of synthetic output and deepfake labels, worked out in the final code of practice of 10 June 2026.

• AI literacy — anyone developing or deploying AI

ensures sufficient knowledge among staff.

International: the frameworks are converging

The AI Act is not the only framework that matters, and it does not stand alone. The international layer runs from binding treaty to voluntary standard, and on the frontier the measures are converging:

• United States — California's frontier AI law (SB 53)

has imposed transparency duties on the largest models since 1 January 2026, while Colorado and Texas open the wider state front.

• Asia-Pacific — South Korea legislates with an

AI Basic Act; the wider region seeks its own balance.

• United Kingdom — a principle-based approach without a

comprehensive statute.

For practice

To find out which of these regimes affect your organisation, take the AI Act self-scan — it maps your systems and gives, per system, the applicable obligations, your gaps and a readiness score, with source references. If you work in a data-intensive chain, the file AI regulation in logistics and transport is the natural starting point.