EU AI Act: what you need to know

Practically summarized — timeline, penalties, who is affected, and the latest deadline changes: transparency & watermarking moved to 2 December 2026, high-risk postponed.

What the AI Act is

An EU regulation adopted in 2024. It regulates AI systems by risk — from prohibited practices, through high-risk systems, to general-purpose AI models. It applies to providers and to deployers — i.e. most EU companies.

Effectiveness timeline

  1. February 2, 2025

    Prohibited practices bans + mandatory AI literacy

  2. August 2, 2025

    Rules for general-purpose AI models and governance structures

  3. August 2, 2026

    General application — a mandatory national AI regulatory sandbox in every member state (Art. 57), governance bodies and the sanctions framework

  4. December 2, 2026

    Transparency & watermarking for AI-generated content (Art. 50) — moved from 2 August 2026, with a 4-month grace period for systems already on the market

  5. August 2, 2027

    Rules for AI inside regulated products (Annex I)

  6. December 2, 2027 (expected)

    High-risk (Annex III) obligations — postponed from 2 August 2026 by the Digital Omnibus (political agreement of 7 May 2026), pending formal EU approval

Recent changes: the transparency & watermarking obligations for AI-generated content (Art. 50) have been formally moved from 2 August 2026 to 2 December 2026, with a 4-month grace period for systems already on the market. The high-risk (Annex III) obligations are being postponed further — under the Digital Omnibus (political agreement of 7 May 2026) they are expected to apply from 2 December 2027, still ahead of formal EU adoption. The 2 August 2026 general-application date still stands for national AI regulatory sandboxes, governance bodies and the sanctions framework.

Penalties

Prohibited practices
up to €35M or 7% of global turnover
High-risk non-compliance
up to €15M or 3% of global turnover
False information to the regulator
up to €7.5M or 1.5% of global turnover

Deployer obligations

  • ·Use high-risk systems per the provider's instructions
  • ·Ensure human oversight appropriate to the use case
  • ·Monitor operation and report incidents
  • ·Inform affected people about AI decisions
  • ·Data governance: quality inputs, DPIA where relevant
  • ·AI literacy across the team (article 4)

The Shadow AI problem

Employees use AI tools that IT and leadership often don't know about. Personal ChatGPT accounts, Copilot outside company IDP, AI features inside SaaS. An auditor asks 'which AI do you use' — and nobody truly knows. Shadow AI is the most frequent source of real incidents.

Overlap with GDPR, DORA, NIS2, ISO 42001

The AI Act is not isolated. Data in AI is still GDPR. Financial firms have DORA. Critical infrastructure has NIS2. ISO 42001 is a voluntary management system covering many of these topics systematically. The GovReady framework is designed so evidence collected once serves multiple frameworks.

EU AI Act — what you need to know · GovReady