EU rail agency adopts AI image labels under AI Act guidance

The European Union Agency for Railways (ERA) has begun introducing labels for AI-generated images on its website, becoming one of the first EU agencies to visibly implement the European Commission’s new transparency guidance under the AI Act.

The European Commission adopted guidelines on 20 July explaining how providers and deployers should comply with the transparency obligations set out in Article 50 of the AI Act. ERA began progressively applying its own measures to new website content the following day.

Images created with generative AI will carry the label ‘AI Generated’, while unlabelled visuals will consist of original ERA photographs, partner-supplied material, commissioned professional work or licensed images.

ERA said synthetic visuals may be used for illustration and communication, but not in ways that could mislead the public about real events, people, locations or factual information. The agency also stressed that it retains full editorial responsibility for everything published on its website.

The labelling policy is intended to help visitors distinguish AI-generated images from authentic or licensed material while allowing the agency to continue using generative AI where appropriate.

Although the AI Act also requires disclosure for certain AI-generated texts concerning matters of public interest, ERA said its written content does not require separate labelling because every publication is subject to human review and editorial oversight.

AI may also support translations where quality is considered sufficiently reliable, while the same editorial standards continue to apply.

By introducing AI-generated image labels, ERA is providing a practical example of how EU institutions can implement the AI Act’s transparency requirements in everyday public communications.

Why does it matter?

ERA’s approach demonstrates how the AI Act’s transparency requirements can be translated into practical organisational policies rather than remaining purely regulatory obligations. Visible labelling, combined with human editorial responsibility, helps public institutions adopt generative AI while maintaining trust in official communications.

The initiative may also provide a model for other public bodies as they develop their own internal AI governance practices. As more organisations begin using generative AI for communications, consistent disclosure policies are likely to become an increasingly important element of public sector transparency.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!