RRT explains AI Act labelling duties and exemptions for professional content

Lithuania’s Communications Regulatory Authority (RRT) says professional users must label specified AI-generated or AI-altered content under transparency requirements that began applying on 2 August 2026.

RRT published its explanation on 19 August. It describes existing requirements under the EU AI Act, rather than a new investigation, fine or general prohibition on AI-generated material. RRT is responsible for supervising compliance with the transparency requirements in Lithuania.

The duty covers companies, public-sector institutions, media outlets, advertising creators and other legal or natural persons using AI content in professional or economic activity. RRT gives commercial social-media influencers as an example of individuals within scope.

Personal, non-professional use is outside the requirement. The dividing line is the use of the content, rather than the user’s organisational form.

Content within scope

AI-generated or altered image, audio and video content must be labelled when it falls within the AI Act’s deepfake category. Publicly released AI-generated or altered text also requires a label when a person has not thoroughly reviewed it and it is intended to inform the public about matters of public interest.

RRT describes public-interest topics as including politics, democratic processes, public administration and services, law enforcement, public safety and health, environmental protection, the economy, finance, science and culture. Text posted on social media is public unless access is limited to a small closed group or the material is distributed through an organisation’s internal communications channels.

Human review can remove the text-labelling duty, but the reviewer must check the facts. Correcting spelling or punctuation alone does not meet the exemption. Text without claims related to socially important topics, such as health, consumer safety or sustainability, is also outside the described scope.

Deepfake labelling is not required for auxiliary editing functions or minor changes that do not alter a person’s perception of the content’s authenticity. RRT gives examples including background edits, lighting adjustments, colour correction, audio parameter changes, noise reduction and file compression.

The label and the sanction

Deepfake content that is obviously part of an artistic, creative, satirical, fictional or comparable work can use simplified labelling. The obligation remains, but the label must not interfere with viewing or enjoyment of the work.

The label must be clearly visible and recognisable from the first contact with the content. RRT says the identifying icon or wording must be integrated directly into the deepfake or text and remain visible when the material is shared or downloaded. It recommends free, standardised EU icons in English.

RRT says the AI Act provides for sanctions of up to €15 million or 3 per cent of a company’s total worldwide annual turnover for breaches of transparency duties. A warning may be possible for a minor infringement. RRT describes the maximum as applying in exceptional circumstances involving particularly serious breaches.

The authority also says the AI Act contains special rules for fines imposed on small and medium-sized enterprises and start-ups. Lithuania’s detailed penalty arrangements will be set by legislation currently under consideration in the Seimas, with amounts expected to reflect the seriousness of the breach, the operator’s income and aggravating or mitigating circumstances.

The requirements took effect on 2 August 2026. RRT’s explanation identifies no separate compliance deadline or scheduled follow-up event; the official notice and the legislation under consideration in the Seimas are the records to monitor.