The EU AI Act: What every editor needs to know
A practical primer on transparency, human oversight, and the questions newsrooms should ask their AI vendors now.
The EU AI Act is not a single rulebook for journalism. It is a risk-based regulation that applies differently depending on what a system does, who deploys it, and where the output is used. That makes it easy to misread in both directions: as irrelevant to editorial tools, or as a ban on AI in newsrooms. Neither is accurate.
This primer is for editors and product leads, not lawyers. It focuses on the questions that affect tool selection and newsroom process.
Who is who
The Act distinguishes providers (who develop or place a system on the market) from deployers (who use it in a professional context). A newsroom using a third-party tool is usually a deployer. A newsroom building its own tool may be a provider.
That distinction matters because obligations attach differently. Deployers have duties around human oversight, monitoring, and, for certain uses, transparency. Providers have duties around documentation, risk management, and conformity assessment.
Risk categories, briefly
- Prohibited practices: a small set of uses, such as social scoring or manipulative techniques that materially distort behaviour.
- High-risk systems: uses listed in the Act, including some in education, employment, essential services, and law enforcement. Most editorial tools are not in this list, but adjacent uses can be.
- Limited-risk systems: transparency obligations, including disclosing that content is AI-generated in certain circumstances.
- Minimal-risk systems: no additional obligations beyond existing law.
Where editorial tools land
An assessment tool that scores article quality and presents evidence to an editor is likely minimal or limited risk, depending on how it is used. A tool that automatically decides what to publish, or that profiles individuals, is a different question.
The Act’s transparency rules matter for synthetic content. If a newsroom publishes AI-generated images or audio, disclosure may be required. Provenance metadata, such as C2PA, is one way to support that disclosure, though it is not a complete compliance answer.
Human oversight is not a slogan
For higher-risk uses, the Act requires effective human oversight. In practice, that means more than a “review” button. It means the reviewer has the competence, information, and authority to override the system, and that overrides are recorded.
Newsrooms should ask vendors:
- What information does the reviewer see?
- Can the reviewer override, and is the override logged?
- What happens when the reviewer disagrees with the model?
- How is reviewer performance monitored?
Transparency obligations for general-purpose AI
The Act also regulates general-purpose AI models, with obligations around documentation, copyright policy, and training-data summaries. Newsrooms are not usually model providers, but the tools they buy may depend on models that are. Ask for the model card and the provider’s compliance statement.
What the Act does not do
- It does not ban AI in journalism.
- It does not create a general right of access to newsroom algorithms.
- It does not replace the GDPR, defamation law, or editorial codes.
- It does not make a vendor’s claim of “AI Act compliant” meaningful without scope and evidence.
Five questions for procurement
- What is the intended purpose? A vague purpose statement makes risk classification impossible.
- Which risk category, and why? Ask for the reasoning, not just the label.
- What human oversight is built in? Ask to see the review interface.
- What is logged, for how long, and can we export it? Auditability is a recurring requirement.
- What changes trigger a new assessment? Model updates can change risk.
The deadline question
The Act applies in stages. Prohibitions and AI literacy obligations began first; general-purpose model obligations followed; high-risk obligations have a longer runway. Deadlines are easy to misstate, so check the current official text rather than a vendor deck.
What to do now
Inventory the AI tools already in the newsroom. Write down what each one does, who reviews its output, and what happens when it is wrong. That inventory is useful for the AI Act, the GDPR, and basic editorial governance. Most newsrooms do not have it.
Compliance is not a certificate you buy. It is a set of practices you can describe and evidence. Start with the description.