AI Act deadline extension: what really applies from 2 august 2026?

Which deadlines are intended to be postponed?

The planned deadline extension primarily concerns the obligations for high-risk AI systems. These obligations are particularly extensive and cover, among other things, risk management, technical documentation, data quality, human oversight, accuracy, robustness and cybersecurity.

Based on the current state of discussion, the following deadlines in particular are expected to be adjusted:

Area	Current status of the planned deadline
High-risk AI systems under Annex III	expected to be 2 December 2027
High-risk AI in regulated products under Annex I	expected to be 2 August 2028
Machine-readable marking of synthetic content under Art. 50(2)	expected to be between November 2026 and February 2027

It is important to note that the planned postponement does not affect all AI Act obligations in the same way. While longer transitional periods are envisaged for high-risk AI systems, the general transparency obligations under Art. 50 AI Act largely remain tied to the previous application date.

For companies, this means that there is likely to be more time for high-risk AI compliance. By contrast, the transparency obligations relating to the use of AI and AI-generated content should still be prepared for by 2 August 2026.

Will the obligation to label AI-generated content also be postponed?

No. The obligation to label AI-generated content is not expected to be postponed.

The postponement currently under discussion primarily concerns the machine-readable marking required under Art. 50(2) AI Act. This refers to the technical obligation of providers of generative AI systems to mark synthetic content such as images, videos, audio or text in a machine-readable format.

For certain existing systems, this technical marking obligation may only start to apply between November 2026 and February 2027. The exact date is still subject to ongoing negotiations.

By contrast, the other central transparency obligations under Art. 50 are not expected to be postponed. These include in particular:

• Chatbot disclosure: users must be able to recognise that they are interacting with an AI system.
• Deepfake disclosure: deceptively realistic AI-generated or manipulated image, audio or video content must be disclosed.
• Information requirements for emotion recognition and biometric categorisation.
• Disclosure of certain AI-generated texts where they are published for the purpose of informing the public about matters of public interest.

For companies, the key point is therefore clear: the obligation to label certain AI-generated content applies from 2 August 2026.

Has the deadline extension already been adopted?

No. The deadline extension has not yet been finally adopted.

The proposed changes are still part of the European legislative process. For a deadline extension to become legally binding, the EU institutions involved must agree on a final text. This text must then be formally adopted and published.

The current situation does suggest that a postponement of the high-risk obligations is highly likely. Nevertheless, until the procedure has been completed, the changes remain proposed amendments rather than final applicable law.

Companies should therefore treat the new deadlines as expected deadlines, but should not assume that all AI Act obligations will automatically apply later.

Why are the deadlines intended to be extended?

The deadline extension is primarily intended to address practical implementation challenges. The requirements for high-risk AI systems are complex and require companies to systematically record, assess, document and monitor their AI systems.

At the same time, many of the practical foundations needed for implementation are still being developed. These include, in particular, technical standards, guidelines, competent authorities, conformity assessment procedures and further implementation guidance.

The extension therefore pursues several objectives:

1. Greater legal certainty for companies

Companies should be better able to assess which specific requirements they need to meet and what implementation should look like in practice.

2. More realistic implementation of the high-risk obligations

The additional months are intended to help companies inventory and classify their AI systems and establish the necessary compliance structures.

3. Avoiding unnecessary bureaucracy

The EU wants to avoid companies having to set up extensive processes that may later need to be adapted again because standards or guidelines have not yet been finalised.

4. Strengthening competitiveness and innovation

The postponement is intended to give companies greater planning certainty and prevent regulatory uncertainty from unnecessarily hindering the development or use of AI.

Conclusion

The planned AI Act deadline extension does not mean that companies can postpone AI compliance altogether. The postponement is intended to apply primarily to the extensive obligations for high-risk AI systems. For these systems, new deadlines of 2 December 2027 and 2 August 2028, respectively, are currently under discussion.

The transparency obligations under Art. 50 AI Act, by contrast, largely remain relevant for 2 August 2026. This applies in particular to information requirements for chatbots, disclosure obligations for deepfakes and certain labelling obligations for AI-generated content.

Only in relation to the technical, machine-readable marking of synthetic content by providers of generative AI systems is a limited transitional period currently being discussed. Based on the current state of discussion, this period ranges from November 2026 to February 2027.