NEW

The EU explains when AI content must be labeled

The European Commission has published new guidelines for the AI Act's transparency requirements. They explain who should label AI content and when human review can provide exceptions for text.

Publicerad 20 July 2026, 20.04

AI-generated illustration of an editor reviewing an image, machine-readable labels and a clear human control point in front of an abstract Europe map.

The European Commission has published guidelines for the transparency requirements in Article 50 of the AI Regulation. They will help providers and businesses that use AI to determine when people need to be told that they are encountering an AI system or seeing AI-generated content.

From August 2, 2026, providers must, among other things, design directly interactive AI systems so that people are informed that they are encountering AI, if it is not already obvious. Providers of generative AI systems must also add effective machine-readable tags that make AI-generated or manipulated content detectable.

Businesses using AI may need to inform people about sentiment analysis, biometric categorization and deepfakes. A machine-readable label alone is not enough for a deepfake. The information must be clear and understandable when the person first encounters the content.

AI-generated or manipulated text on matters of public interest should be flagged when published without human review or editorial control. The Commission explains that a simple spelling or grammar check is not enough. The reviewer needs to assess the content in substance and be able to change or stop the publication.

The guidelines also contain exceptions and examples. Regular editing, source code and some closed industrial workflows may fall outside the labeling requirement. Exact liability depends on the actor's role, the content and how it is used.

The rules apply from August 2 without any general respite. A limited exception applies to labeling and discoverability for AI systems released before this date. For them, that part applies from 2 December 2026. Older content does not need to be marked afterwards.

Därför spelar det roll

Swedish companies, authorities, media and agencies now receive more concrete guidance ahead of August 2. The practical difference is that responsibility doesn't just lie with whoever builds the AI system. Even the business that uses AI for publishing, deepfakes or systems that assess emotions may need to inform people clearly.

Det här kan du göra

  1. Map where the business uses AI in customer contact, publishing, image, audio and video.
  2. Document who fact-checks, can change the content and is responsible for the final publication.
  3. Check the full guidelines for each use case. This article is not legal advice.