A few weeks ago an AI-generated image of a shoe attributed to Lannert Judit circulated widely across Hungarian media and social networks, and was even cited in various expert analyses. The episode underscored how rapidly generative artificial intelligence can produce visual and textual content that enters public discourse and is treated as if it were real.
Trends in technology and regulation
Generative AI is advancing toward ever more photorealistic outputs; the next major challenge is likely to be real-time, photorealistic video. As a result, provenance and authenticity checks are becoming more important because visual inspection alone will increasingly fail to distinguish real from synthetic content.
Regulatory responses are already in place: the EU AI Act has applied since August 2, 2026, and requires machine-readable labels for certain AI-generated or manipulated content and audience-visible notices for deepfakes and some public-interest materials.
Industry actors are also taking steps. Anthropic has announced watermarking for content produced by Claude. At the same time, industry standards for reliably identifying text and visual AI outputs remain under development.
Technical limits of watermarking and provenance tools
Watermarks and provenance technologies can provide useful signals for everyday users, since most people will not deliberately try to remove or evade such markers. Nevertheless, these methods have clear limits: embedded markers and metadata can be stripped or altered, and text-based markers can be obfuscated by rewriting. Consequently, watermarking is a valuable tool but not a foolproof defense against deliberate disinformation campaigns.
Source criticism and the centrality of trust
Everyday information consumption usually does not allow thorough verification of every item: users encounter hundreds of images and videos and thousands of lines of text daily. Therefore, who or what sources we trust will remain decisive. Research from the U.S. National Institute of Standards and Technology (NIST) highlights that trust depends not only on system performance but also on design elements that help users interpret performance — in other words, transparency is key to understanding how a piece of content was produced.
The goal of transparency is to set realistic expectations rather than stigmatize — human-created content can be misleading, and machine-generated analysis can be accurate and valuable.
Consumer expectations and corporate choices
Consumers increasingly expect disclosure: in a 2026 Capgemini survey of 12,000 consumers, 67% said they expect companies to clearly indicate when an advertisement shown to them was AI-generated. For firms, this means creating clear, consistent internal rules about AI use and deciding when to tell audiences about AI’s role.
Beyond legal requirements, disclosure is an ethical and communications decision for companies: they must decide when transparency offers a business advantage or when it is simply the correct approach toward their audience.
Practical takeaways
- It is reasonable to label AI-generated images, videos or text when they are presented directly to users. When AI operates in the background — for example in recommender systems — it may not be necessary to tag every piece of content.
- Watermarks and provenance are useful but not infallible; motivated bad actors can remove or obscure such signals.
- Because routine, full verification is impractical, source-checking and the reputation of publishers, journalists, experts and even acquaintances will become more important in judging credibility.
Author and events
This piece was written by Kiss Gergely, co-founder and president of Attrecto Zrt. and a board member of the Signifer Group. The topic will also be discussed at the Portfolio AI & Digital Transformation conference on November 26, 2026.
Cover image: illustration. Tags: media, transparency, technology, artificial intelligence, forgery, trust, deepfake, disinformation, generative AI.
