Back

EU AI Act Implementation Progresses Amid Delays and Uncertainty

lite News Desk 2 min read

At a glance

  • The EU AI Act entered into force on 1 August 2024.
  • Transparency rules became enforceable on 2 August 2026.
  • High-risk AI obligations were postponed to late 2027 and 2028.

The EU AI Act established a risk-based legal framework for artificial intelligence, introducing phased obligations for transparency and high-risk systems. Recent developments highlight ongoing challenges in aligning technical standards and governance structures across Member States.

Transparency measures, including requirements for disclosing AI interactions and applying machine-readable watermarks to synthetic content, became enforceable on 2 August 2026. These provisions aim to increase clarity for users interacting with AI-generated material and set out obligations for providers of such systems.

Obligations for high-risk AI systems, such as those used in biometrics, employment, medical diagnosis, and critical infrastructure, were postponed following the entry into force of the Digital Omnibus amendments on 27 July 2026. The new deadlines set compliance dates for December 2027, with an extension to August 2028 for AI embedded in products.

Member States are required to establish or participate in at least one AI regulatory sandbox to facilitate compliance testing in controlled environments. However, researchers have identified issues related to the design of these sandboxes, as well as fragmentation and timing challenges that affect their effectiveness.

What the numbers show

  • The EU AI Act became law on 1 August 2024.
  • Transparency requirements took effect on 2 August 2026.
  • High-risk AI system obligations are delayed to December 2027 and August 2028.

The governance structure for the AI Act includes the AI Office, national competent authorities, and the European Artificial Intelligence Board, with support from advisory groups. Member States were expected to designate national authorities by 2 August 2025, but many did not meet this deadline, contributing to implementation delays.

Stakeholders participating in Commission consultations and a “reality check” event in September 2025 identified several challenges, including the absence of harmonised standards, limited guidance, a lack of notified bodies, and uncertainty regarding the scope of the rules and governance fragmentation.

Scholars have described the AI Act as a form of conditional regulation, noting that enforceability depends on technical standards that have not yet been fully defined. This situation has led to legal uncertainty for those subject to the regulation.

Implementation of the AI Act is currently being tested in practice, with ongoing analysis of compliance burdens for organisations of different sizes and sectors. Questions remain about the readiness of various industries and whether the regulatory framework will achieve its intended objectives.

Industry reaction

Anthropic announced that future versions of its Claude models will include watermarking features to meet the transparency requirements set out in the AI Act.

* This article is based on publicly available information at the time of writing.

More from EU AI Regulation

  1. New EU AI Act rules require machine-readable marks on AI-generated content starting August 2026, with fines up to €15 million for non-compliance.

  2. The EU's AI Act enforcement began on 2 August 2026, mandating AI content labeling and imposing fines up to 7% for violations, according to reports.

  3. A new enforcement action began on 2 August 2026, allowing fines up to €15 million for non-compliance, according to the regulator.

  4. The European Commission unveiled an Action Plan for AI and cybersecurity. It aims to establish seven AI Gigafactories by 2035, according to reports.