Skip to content

The EU AI Act in 2026

What already applies, what was postponed, and what to do next

The EU's “Digital Omnibus” pushed the high-risk deadlines further out and has been in force since 27 July 2026. The Article 50 transparency and labelling obligations, by contrast, have applied since 2 August 2026. This guide separates what is binding already, which deadlines are still ahead, and what you have more time for.

As of October 2026. Not legal advice.

Where things stand in October 2026

The EU AI Act (Regulation (EU) 2024/1689) has been in force since August 2024 and applies in phases. The “Digital Omnibus”, the reform package that significantly delays the obligations for high-risk AI, is now law: Parliament approved it on 16 June 2026 and the Council on 29 June 2026 (Council press release of 29 June 2026). It was published as Regulation (EU) 2026/1744 of 8 July 2026 in the Official Journal of the EU on 24 July 2026 (EUR-Lex) and has been in force since 27 July 2026. This is not the all-clear: several duties are binding already, and the Article 50 transparency obligations have applied since 2 August 2026. They affect every company that runs chatbots or generative AI.

  • Prohibited AI practices have been banned since 2 February 2025, including social scoring, manipulative systems and untargeted scraping of facial images.
  • The AI-literacy duty (Article 4) has applied since February 2025. The Omnibus reworded it: providers and deployers must take measures that support their staff's AI literacy. Article 4 no longer requires a specific level of literacy for each person.
  • Obligations for general-purpose AI models, the large language models, have applied since 2 August 2025.
  • Since 2 August 2026 the Article 50 transparency obligations have applied: chatbots must disclose that they are AI, and AI-generated content and deepfakes must be labelled.
  • The high-risk obligations have been postponed, firmly to 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
  • In Germany, the AI Market Surveillance Act (KI-MIG) is in force (Federal Law Gazette 2026 I No. 223 of 28 July 2026). The central market-surveillance authority is the Bundesnetzagentur.

Deadlines: in force vs. postponed

The AI Act does not land all at once; it phases in. With the Omnibus (Regulation (EU) 2026/1744), the roadmap looks like this:

Key dates:

  • Since 2 February 2025: ban on unacceptable AI practices (Art. 5) and the AI-literacy duty (Art. 4, reworded by the Omnibus since 27 July 2026).
  • Since 2 August 2025: obligations for GPAI models, the governance structure (AI Office, national authorities) and the penalty regime.
  • Since 2 August 2026: Article 50 transparency obligations, i.e. chatbot disclosure, machine-readable marking of AI-generated content, labelling of deepfakes and AI-generated text on matters of public interest.
  • From 2 December 2026: the new Omnibus ban on AI systems that generate non-consensual intimate imagery or abuse material (Art. 5). The same day, the transitional period for machine-readable marking of existing systems ends.
  • From 2 December 2027: obligations for stand-alone high-risk systems under Annex III (moved from August 2026).
  • From 2 August 2028: obligations for high-risk AI embedded in already-regulated products under Annex I (e.g. medical devices, machinery).

Transitional rule for marking: generative AI systems that were already on the market before 2 August 2026 have until 2 December 2026 to implement machine-readable marking of their outputs. All other Article 50 obligations have applied since 2 August 2026 without deferral. Source: Regulation (EU) 2026/1744, Official Journal of the EU of 24 July 2026.

The four risk classes

The AI Act does not regulate “AI” as a blanket category — it regulates by risk. Which class your system falls into determines your obligations:

The four tiers at a glance:

  • Unacceptable risk — prohibited. Social scoring, manipulative or exploitative systems, mass biometric surveillance. Banned since February 2025.
  • High risk — strictly regulated. AI in hiring, credit scoring, critical infrastructure, education or medical devices. Requires risk management, data governance, human oversight and documentation. Deadline postponed (see above).
  • Limited risk — transparency duty. Since 2 August 2026, chatbots, AI-generated content and deepfakes must be labelled as such (Article 50).
  • Minimal risk — unrestricted. The bulk of everyday business AI (spam filters, recommendations, productivity tools). No specific duties; voluntary codes of conduct encouraged.

Provider or deployer: who is liable for what

The AI Act distinguishes between the “provider” (who develops an AI system or places it on the market under their own name) and the “deployer” (who uses an AI system in a professional capacity). Most mid-market companies are deployers — and often underestimate that this role carries duties too. But adapt a system substantially, or pass it on under your own brand, and you can become a provider yourself.

As a provider

  • Conformity assessment and CE marking for high-risk systems
  • Technical documentation and a risk-management system
  • Registration in the EU database before going to market
  • Post-market monitoring

As a deployer

  • Use in line with the instructions and ensuring human oversight
  • Measures for the AI literacy of the staff who use the system (already mandatory)
  • Transparency towards affected individuals; informing the workforce
  • For high-risk use: a fundamental-rights impact assessment where required

Establish your role per system first — it drives the entire set of obligations. When in doubt, for instance when fine-tuning or white-labelling third-party models, check early whether you cross over into being a provider.

What to do now

1Take inventory

  • Build a register of every AI system in use — including features embedded in SaaS tools
  • Determine the risk class for each system
  • Define your role for each system: provider or deployer

2What counts immediately

  • Check whether any prohibited practices are in use — these must stop now
  • Set up AI-literacy measures for the staff involved (already mandatory)
  • Implement transparency notices for chatbots, AI-generated content and deepfakes: mandatory since 2 August 2026

3Use the runway

  • For high-risk systems, build risk management and documentation now rather than waiting until 2027
  • Structure data governance and traceability early
  • Map your supply chain — which models and components come from third parties?

4Embed governance

  • Name an owner for AI compliance
  • Connect it to the GDPR — many requirements overlap
  • Know your regulator: in Germany, the Bundesnetzagentur is the central AI market-surveillance authority (AI Market Surveillance Act)

Frequently asked questions about the EU AI Act

My company only uses ChatGPT and Microsoft Copilot. Am I affected?

Yes, as a deployer. Even if you only use AI rather than build it, the AI-literacy duty and transparency rules apply. On top of that, using such cloud tools overlaps with the GDPR the moment personal data flows into your prompts.

Has the EU AI Act now been scrapped or postponed?

Neither scrapped nor entirely postponed. The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) moves the high-risk obligations to 2027/2028. The existing bans and GPAI rules remain; the AI-literacy duty has been reworded, not removed. The Article 50 transparency obligations have applied as planned since 2 August 2026.

What exactly has applied since 2 August 2026?

The Article 50 transparency obligations: chatbots must disclose that they are AI, generative systems must mark their outputs in a machine-readable way, and deepfakes as well as AI-generated text on matters of public interest must be visibly labelled. The only exception: generative systems that were already on the market before that date have until 2 December 2026 for the machine-readable marking.

What does a violation cost?

For prohibited practices, up to EUR 35 million or 7% of global annual turnover. For breaches of other obligations, up to EUR 15 million or 3%. For supplying incorrect information to authorities, up to EUR 7.5 million or 1%.

Should the delay make me wait on high-risk compliance?

No. Building risk management, documentation and data governance takes months. The extended deadline is a buffer, not a reason to defer — and most of the work pays into GDPR and IT security anyway.

How do the AI Act and the GDPR fit together?

They interlock. As soon as AI processes personal data, both apply. A clean GDPR foundation — legal basis, data minimisation, data-subject rights — is often half the battle for AI Act conformity. Our GDPR-compliant-AI guide complements this overview.

Related resources

Not sure which obligations apply to your AI systems?

We map your AI landscape into the risk classes, clarify your role as provider or deployer, and show which steps matter now, GDPR-compliant and without the hype.

Or start with our structured AI potential and compliance analysis.

Go to the AI readiness analysis