Entry Into Force

Legal Framework

The temporal scope of the AI Act is governed primarily by Article 113, which establishes a two-stage architecture: entry into force (the Regulation becoming law) and application (obligations becoming enforceable). The distinction is critical for compliance planning.

"This Regulation shall enter into force on the twentieth day following that of its publication in the Official Journal of the European Union. It shall apply from 2 August 2026. However:"
— AI Act Art. 113

The word "However" introduces a staggered application schedule: certain provisions—particularly the prohibited practices under Article 5—became applicable earlier, while obligations for high-risk AI systems under Article 3 definitions follow a longer runway. The general application date of 2 August 2026 applies to most substantive obligations, including those binding providers, deployers, importers, and distributors as defined in Article 3.

Separately, Article 112 establishes a review mechanism that is itself temporally anchored to entry into force:

"The Commission shall assess the need for amendment of the list set out in Annex III and of the list of prohibited AI practices laid down in Article 5, once a year following the entry into force of this Regulation"
— AI Act Art. 112(1)

This means the Commission's annual review obligation has been running since the Regulation's entry into force, even before most provisions became applicable.

Key Developments

The EDPB and EDPS, in their 2021 Joint Opinion on the original proposal, flagged concerns about the implementation timeline—particularly regarding legacy systems:

This position shaped the eventual compromise on transitional provisions. More recently, in early 2026, the Commission proposed targeted amendments to simplify implementation:

The EDPB and EDPS responded by cautioning against relaxing obligations such as AI literacy and opposing excessive postponement of high-risk rules, arguing that providers and deployers should not be released from core obligations during any transitional period.

Status of the Debate

This topic is developing. No dominant doctrinal pattern has emerged yet, and the landscape remains in flux. The Commission's 2026 simplification proposal has introduced active debate over whether the staggered application timeline should be further extended, particularly for high-risk AI systems. The EDPB-EDPS position signals resistance to broad deferrals, but the final legislative outcome is unresolved. No court has yet interpreted Article 113's transitional provisions. The open question—whether the staggered deadlines adequately balance innovation with protection—will likely be settled through legislative amendment rather than litigation, though enforcement actions after 2 August 2026 may clarify the boundaries.

Practical Guidance

• Map obligations to specific application dates. Not all provisions apply on 2 August 2026. Identify which obligations govern your AI systems—prohibited practices, GPAI models, high-risk systems—and trace each to its applicable date under Article 113's staggered schedule.
• Audit existing systems against transitional provisions. AI systems already placed on the market before application dates may benefit from limited transitional relief, but the EDPB-EDPS position suggests narrow interpretation. Do not assume blanket grandfathering.
• Monitor the 2026 simplification proposal. The Commission's proposed amendments to the implementation timeline for high-risk rules could shift compliance deadlines. Track legislative progress closely, as changes may affect readiness planning.
• Prepare AI literacy programmes now. The EDPB and EDPS have signalled that AI literacy obligations should apply without postponement. Ensure staff training is operational before the relevant application date rather than awaiting potential extensions.
• Track the Commission's annual review under Article 112. The annual assessment of Annex III and prohibited practices means the regulatory perimeter can shift yearly. Build a monitoring process to detect additions that could reclassify your systems as high-risk.