What the guidelines cover
Since 2 August 2026, the transparency obligations under Article 50 of the EU AI Act apply to organisations in the EU. According to BornCity, companies and deployers must expressly inform users that they are interacting with an AI system and clearly label AI-generated or manipulated content. To clarify implementation, the European Commission published guidelines in the Official Journal on 12 November 2026 under number C/2026/6005.
According to the source, the guidelines are not legally binding, as the final interpretation rests with the European Court of Justice. They structure the obligations under Article 50 of Regulation (EU) 2024/1689 into four areas.
The four obligations
For interactive systems, providers must design interfaces so that users can unmistakably recognise AI use. The only exception applies when this is completely obvious to an average consumer. Synthetic content must be marked in a machine-readable way. A mere marking is not sufficient; technical solutions should be effective, interoperable, robust and reliable as far as technically feasible.
Deployers of emotion recognition or biometric categorisation systems must inform all exposed persons, including minors, in advance. Deployers must also disclose deepfakes and artificially generated text on matters of public interest. For text, an exception applies if substantial human review by professionals takes place and editorial responsibility is assumed. According to the source, spelling corrections, general editorial guidelines or automated review procedures are not enough. If a text is reworded with AI after approval, the exception lapses entirely.
Deepfakes and advertising images
According to BornCity, the definition of deepfakes under Article 3 point 60 rests on four cumulative criteria, including recognisable resemblance to real or plausible persons, objects, places or events. Intent to deceive is not required; the real audience, including vulnerable groups, is decisive. For product images, a manipulated picture counts as a deepfake if it can mislead about appearance, properties or benefits. Mere colour correction or background replacement remains exempt, as long as the product depiction is not misleading.
According to the source, an invisible watermark does not replace a visible label: marking must be perceptible without additional tools. Companies outside the EU may also be covered if they publish deepfakes on the open internet and an EU audience is foreseeable.
Deadlines and penalties
The Omnibus Regulation (EU) 2026/1744 of 24 July 2026 grants providers of generative systems that were on the market before 2 August a transitional period for machine-readable marking until 2 December 2026. Further obligations for high-risk systems apply from 2 December 2027 and 2 August 2028 respectively. According to BornCity, violations can lead to fines of up to 15 million euros or 3 percent of worldwide annual turnover, with the lower value applying to SMEs and start-ups.
Assessment
For companies using generative systems, the clarification could mainly settle whether a label must be visible and when the editorial exception applies. Whether the guidelines suffice in practice depends, according to the source, on later interpretation by the CJEU, which is binding.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
