A four‑month transitional period for AI systems that were already available in the European Union expires on December 2, 2026. From that date, operators of AI systems that were accessible in the EU before August 2, 2026 must comply with the EU AI Act’s transparency requirements for AI‑generated or significantly modified content.
According to EY (Ernst & Young), the change affects not only large international technology companies but also Hungarian software developers, businesses that integrate AI functions into their products, and firms that sell or use third‑party AI solutions under their own brand.
What the regulation requires
The EU regulation aims to make it identifiable to both people and digital systems when content was produced or substantially modified with the help of artificial intelligence. The obligation covers text, images, audio and video. Origin marking can be implemented in several ways: embedded file information, metadata, digital watermarks or other machine‑readable signals. The law does not mandate a single technological solution but requires that the chosen method reliably supports identification of AI‑created content.
Markings must survive later transformations
Compliance goes beyond inserting technical markers into AI content: organisations must ensure that origin information persists during later uses. File downloads, compression, resizing, format conversion or uploading to different platforms can remove metadata or watermarks, so companies should test that their chosen markers remain intact after such operations.
Who is affected?
Specific obligations depend on the role a company plays in providing or using a given AI solution. A single organisation can be both a user and a provider in different contexts. Zala Mihály, partner responsible for EY’s IT and Technology Advisory services, warned that because AI is increasingly embedded into corporate software and services, a Hungarian company can acquire provider obligations even when it integrates another developer’s AI technology into a product marketed under its own name.
He also emphasised that preparing for compliance is not solely the task of IT: legal, procurement, marketing, communications and content teams may need to coordinate, particularly where organisations process or publish material originating from external AI providers.
Machine markings are not the same as visible user labels
The December 2 deadline primarily concerns machine‑readable identification of AI‑generated or modified content; it does not automatically require a visible label on every such item. However, in certain situations — notably when publishing deepfakes, i.e. realistic AI‑created or altered image, audio or video content that could mislead about real persons or events — organisations must directly inform users. For AI‑produced text, visible disclosure depends on factors such as the purpose of publication, the AI’s role in creation, the extent of human oversight and whether an individual or organisation accepts editorial responsibility for the content.
Why prepare in advance?
The aim of the new requirements is not to restrict AI use but to improve transparency so people, companies and online systems can identify AI‑created or altered content more easily. Proper preparation helps ensure legal compliance and can contribute to preserving digital trust and protecting corporate reputation, the EY partner noted.
Hungarian companies should review which AI solutions they offer or use, whether some are only for internal purposes, and how they handle marking of generated content — with particular attention to ensuring technical markers remain robust against common file operations.
