In short
An AI policy sets out what the newsroom allows, what it forbids and who approves. It covers human review before publication, disclosure to readers, source protection and records of where texts come from. It helps a newsroom organise the editorial exception in Article 50 of the AI Act. According to the European Commission’s guidelines, merely having a policy is not enough to qualify for this exception. Each text must actually be reviewed.
Many newsrooms have adopted an AI policy since 2023. Since 2 August 2026, the AI Act has made these documents legally relevant. This article sets out what a policy should cover and what the regulation requires.
The clauses of a policy
A useful policy answers specific questions:
- which uses of AI are allowed, and which are forbidden
- who reviews a text before publication, and what they check
- which person or role holds editorial responsibility
- when and how the newsroom informs readers
- which data must never be given to an external tool
- which tools are allowed, and on what terms
- how the newsroom keeps a record of where texts come from
- when the policy is revised
What the AI Act requires
Since 2 August 2026, Article 50 of the AI Act has required disclosure that a text on matters of public interest has been generated or manipulated by AI. The obligation does not apply if the text has undergone human review or editorial control, and if a natural or legal person holds editorial responsibility for its publication.
The European Commission’s guidelines set out these conditions:
- human review covers the substance of the text and is carried out by a person with expertise in the field
- checking the facts is the minimum
- editorial control is exercised by a responsible editorial entity, such as an editor-in-chief
- editorial responsibility can lie with a person, an editorial board or the publishing company
- the identity and contact details of whoever holds it must be easy to find, for example in the legal notice
- a spelling or grammar correction, the mere existence of an editorial policy, an automated review or a cursory approval is not enough
- if AI modifies, adds to or rephrases the text after approval, the exception no longer applies
These guidelines were approved on 20 July 2026. They are not binding.
A starting template
Adapt these clauses to each newsroom:
- The journalist who signs a text answers for it, including passages produced with AI.
- No text produced or modified by AI is published without a human review of its substance, including fact-checking.
- The editor-in-chief holds editorial responsibility for published content. Their contact details appear in the legal notice.
- Quotes are reproduced word for word. AI never rephrases them.
- No source’s identity and no confidential document is given to a tool whose hosting and terms of use have not been approved.
- A text modified by AI after approval goes back for approval.
- The newsroom informs readers about deepfakes and about AI-generated texts published without human review.
- The policy is revised every year.
Reference texts
Reporters Without Borders (RSF) and sixteen partner organisations published the Paris Charter on AI and Journalism on 10 November 2023. It sets out ten principles. They include the media’s responsibility for all published content, transparency about the use of AI, and the traceability of where content comes from. In France, the Charter of Professional Ethics for Journalists and the Munich Declaration remain the reference texts on journalism ethics.
Applying the policy
The editorial exception requires each text to be reviewed. The newsroom must be able to show that this review took place. Verbatim records the origin of every line of an article: the journalist, the AI or a source. The record is ready when the AI Act requires it. Every claim keeps the excerpt that proves it and its verification status. In VerbatimOS, the article goes through the approval queue before publication, and the AI Act report covers the whole team.
Questions
Is an AI policy mandatory?
No law requires one. Article 50 of the AI Act requires a disclosure on AI-generated or AI-manipulated texts on matters of public interest, unless the editorial exception applies. A policy helps organise the human review of each text.
Is a policy enough to qualify for the editorial exception?
No. According to the Commission’s guidelines, merely having an editorial policy is not enough. Each text must undergo human review of its substance or editorial control.
Does the policy have to be published?
The regulation does not require it. According to the guidelines, however, the identity and contact details of the person or entity holding editorial responsibility must be easy to find.
Where should a newsroom start?
List the AI tools already in use in the newsroom and what each desk uses them for. Then write the rules based on that actual use. Verbatim’s AI disclosure generator helps write the sentence shown to readers.