In August 2026, a number of key provisions from the European AI Regulation came into force. Users must now be able to clearly recognise when they are dealing with artificially generated or altered content. The new requirements are initially aimed in particular at providers of AI systems, such as OpenAI, Google and Anthropic. In future, AI-generated content must be labelled in such a way that it can be recognised as artificially generated by machines.
Chatbots and voicebots, such as those used in customer service, must also be clearly recognisable as such to users. Special transparency obligations also apply to so-called deepfakes: if deceptively realistic images, videos or audio recordings are published that artificially depict or significantly alter real people, events or situations, companies must now label them as such. The same applies to AI-generated news articles published without human verification.
“The AI Regulation is intended to ensure that we can tell whether a text, an image or a review has been produced by a human or by artificial intelligence,” summarises MEP Hannes Heide (S&D Group).
Labelling: Yes! But exactly how is not regulated
In practical terms, this means that businesses must label content such as videos, images or texts that have been created or significantly altered using AI. However, there are no specific guidelines on exactly how this should be done.
The European Commission’s practical guide recommends using clearly visible labels such as ‘AI-generated’ or ‘AI-modified’. For videos and audio files, the label should be visible at the start and should be displayed continuously or at regular intervals, or repeated. TikTok and music streaming platforms are already using labels for AI-generated content and music.
Strict regulations for AI in the workplace
The S&D Group in the European Parliament has helped to enshrine the protection of fundamental and workers’ rights in the AI Directive. One safeguard for workers is that companies must inform and involve employee representatives and trade unions before deploying high-risk AI systems in the workplace.
One example is the strict requirements for recruitment processes. If AI is used to assist with staff selection, this must take place under human supervision. Applicants must be informed of this.
Austria is lagging behind in implementation
It remains to be seen how rigorously these transparency rules will actually be monitored and how strictly breaches will be penalised. As with the General Data Protection Regulation, only practical experience will reveal how frequently sanctions are imposed.
Austria is currently lagging behind in implementation. The so-called implementing legislation is still lacking, as is a clearly designated supervisory authority. Nor is there yet a body responsible for penalising breaches
In any case, the aim of the new laws is not to restrict the use of AI, but to build trust, make manipulation more difficult and clarify responsibilities. Breaches can have serious consequences: depending on the nature of the breach, fines of up to 15 million euros or three per cent of global annual turnover may be imposed.
Are private users also liable?
Not all private use is subject to the labelling requirements. For example, anyone who publishes an AI-generated image on their own private profile is not, in principle, required to label it. There are also exceptions for artistic, satirical, fictional or creative works. In these cases, the label does not have to be affixed directly to the work itself if this would detract from its artistic design.
Labelling is important because the naked eye often fails to recognise AI as such
The issue of labelling becomes all the more important the more difficult it becomes to identify AI-generated content retrospectively. Applications are becoming increasingly sophisticated. Images, videos, voices and texts appear ever more realistic. This makes retrospective identification difficult. That is why regulation begins at the stage of content creation and processing.
This does not ban artificial intelligence, but it does make it more transparent. It is crucial that it remains clear when AI has been used, who is responsible for it, and which decisions were ultimately made by humans.
