EU AI Transparency Rules from 2 August 2026: What Companies Must Label — and What They Do Not

New AI transparency obligations are about to become enforceable

From 2 August 2026, the transparency obligations under Article 50 of the EU AI Act begin to apply. Together with the European Commission's final Guidelines and Code of Practice, businesses now have much greater clarity on when AI-generated content must be disclosed—and when it does not.

The potential consequences are significant. Non-compliance with the AI Act may lead to administrative fines of up to €15 million or 3% of worldwide annual turnover, depending on the applicable infringement.

The good news is that many organisations have misunderstood the rules. Contrary to popular belief, not every AI-assisted article, report or image requires an "AI-generated" label.

The biggest misconception

Many companies have started placing "Generated by AI" notices on every publication.

This is not what the AI Act requires.

For legal analyses, policy papers, research articles and other publications informing the public on matters of public interest, disclosure is generally not required where all of the following conditions are met:

• the content has undergone substantive human review;

• there is genuine editorial control;

• a specifically identifiable person accepts legal and editorial responsibility for publication.

Simply correcting grammar or reading the text once before publishing is not sufficient.

Human review must include verification of facts, legal sources, reasoning, recommendations and conclusions.

Editorial responsibility becomes a compliance requirement

The new guidance effectively transforms editorial governance into a compliance issue.

Organisations relying on AI should be able to demonstrate that:

• authoritative sources were verified;

• legal or technical conclusions were reviewed;

• recommendations were assessed by a qualified reviewer;

• the final version was approved by a responsible editor.

Maintaining this evidence may become critical if transparency obligations are later questioned.

What about AI-generated images?

Not every AI-generated image requires disclosure.

Illustrations, conceptual graphics and abstract visualisations generally do not qualify as deepfakes merely because they were created using generative AI.

However, disclosure is required where an image realistically depicts or imitates:

• an identifiable person;

• a real event;

• an existing place;

• an authentic-looking document;

and the image could reasonably be mistaken for genuine.

In such cases, the disclosure should appear no later than the first time the content is displayed.

Practical actions before 2 August

Organisations publishing AI-assisted content should review their editorial procedures before the rules become applicable.

Recommended actions include:

• appoint a responsible editor with final approval authority;

• introduce documented fact-checking and source verification;

• maintain an editorial evidence log;

• adopt an internal AI drafting policy;

• avoid automatically labelling every publication as AI-generated without legal necessity.

A practical editorial statement could read:

AI may assist drafting; final legal analysis, source validation and editorial responsibility remain with Corvus.

A new compliance opportunity

The new rules also create a commercial opportunity.

Many organisations still lack clear procedures for AI-generated publications.

A structured AI Content Transparency Review can help businesses:

• classify AI-generated content;

• determine when disclosure is legally required;

• establish editorial review procedures;

• document human oversight;

• prepare evidence for regulatory inspections.

For regulated sectors, this is rapidly becoming part of broader AI governance rather than simply a communications issue.

Key takeaway

The AI Act does not require organisations to label every piece of AI-assisted content.

Instead, it requires organisations to demonstrate meaningful human oversight where disclosure exemptions apply.

For legal, regulatory and policy publications, robust editorial governance may become just as important as the technology itself.

Need to assess whether your organisation's AI-generated content complies with the new EU transparency rules?

CorvusAI Legal helps organisations design AI content governance, editorial review procedures and transparency frameworks aligned with the EU AI Act before enforcement begins.