Copyright © Media Law International 2026. All Rights Reserved. Privacy Policy Contact Us Unsubscribe
Guide to the Global Leaders in Media Law Practice
REGIONS WE COVER
Middle East and North Africa
North America
Asia-Pacific
Western Europe
Central and Eastern Europe
These values may be summarised under the common denominator of “epistemic justice”, understood as the ethical call to preserve the conditions for a shared reality. Without transparency about AI content generation, the risk being run is what has been termed “truth decay”, that is the “erosion of the role of facts and data in public discourse”, characterised by “growing disagreement about facts and their interpretation” and “a blurred distinction between opinion and fact”.
The risks involved in the undisclosed usage of AI are amplified by the inherent credibility that AI systems (such as LLMs) are capable of obtaining in the eyes of the target audience. As has been observed by prominent scholars (such as Quattrociocchi of Rome La Sapienza University), the cleanliness, fluency and plausibility of the output, and the sense of completeness and quality that AI‑generated output conveys, function as a “cognitive stopping rule”, lowering the cost of doubt and making critical reasoning “quietly unnecessary, as answers arrive before questions, coherence precedes understanding”.
Consider the scenario, which increasingly reflects the current reality of access to “information”, in which content created by AI floods the infosphere to such an extent that it quantitatively surpasses human‑generated content. The widespread dissemination across media outlets of AI‑generated content is incentivised by the cost/benefit ratio: in a context where the currency is users’ attention, even more than or alongside users’ personal data, the speed with which plausible and coherent content is generated by AI, at a fraction of the cost of having it human‑generated, is a strong incentive for media outlets to use AI systems for content generation.
In turn, this triggers a vicious cycle, whereby end‑users lose touch with the inherent value of content, as content scarcity is no longer an issue, while content quality is blurred by the mere resemblance of quality. End‑users who do not know that content has been AI generated are thereby deprived of agency, empowerment and critical thinking. This eventually leads to epistemic injustice.
The above appears to depict a somewhat dystopian scenario, which may be explained by the first prong of Amara’s Law (named after futurist Roy Amara), according to which “we tend to overestimate the effect of a technology in the short run”. However, this law continues: “…and underestimate the effect in the long run”.
Underestimation is, I think, a function of habit: once the adoption of a given technology has become mainstream, end‑users will already have conformed their way of interacting with the technology to the behavioural code dictated by the design of the technology itself. In other words, in the long run, underestimating the effect of a given technology is the consequence of the intrinsic normative value of technology design which, once the tipping point of adoption has been reached, becomes indistinguishable from the norm of social behaviour, which in turn will be informed primarily by the technology itself.
The European Union’s approach to regulating (new) technologies has often been criticised for being overly prescriptive and for adding red tape for businesses that want to adopt them. One example is the safeguards applicable to personal data processing, implemented through legal provisions with the aim of empowering users and providing them with the choice of disentangling themselves from digital technologies’ default feature of being sticky, i.e. their capability of recording and monitoring all that users do online to an unimaginable level of granularity and from there even predicting with precision individuals’ behaviour or inferring their personal conditions (as is known, according to S. Zuboff this results in “surveillance capitalism”).
In retrospect, however, the EU approach should be reappraised and praised, for it represents an advanced attempt to democratise technology, i.e. to rebalance the above‑mentioned innate (by design) capability of technologies to dictate and shape human behaviours and social norms, with the pursuit and protection of other values. As mentioned, these values are the democratic process, epistemic justice, self‑agency and empowerment.
Recital 133 of the EU AI Act fits quite aptly into the above‑described digital reality, stating that “A variety of AI systems can generate large quantities of synthetic content that becomes increasingly hard for humans to distinguish from human‑generated and authentic content. The wide availability and increasing capabilities of those systems have a significant impact on the integrity and trust in the information ecosystem raising new risks of misinformation and manipulation at scale, fraud, impersonation and consumer deception”.
The impact of the concealed use of AI systems in content generation is thus identified in Recital 133, which then also sets out the interpretative framework concerning the relevant remedies. In this regard, Recital 133 calls on providers of AI systems to “embed technical solutions that enable marking in a machine readable format and detection that the output has been generated or manipulated by an AI system and not a human. Such techniques and methods should be sufficiently reliable, interoperable, effective and robust as far as this is technically feasible, taking into account available techniques or a combination of such techniques, such as watermarks, metadata identifications, cryptographic methods for proving provenance and authenticity of content, logging methods, fingerprints or other techniques, as may be appropriate”.
Recital 133 further qualifies the adoption of such remedies by providing that, when discharging this obligation, providers should also take into account “the relevant technological and market developments in the field, as reflected in the generally acknowledged state of the art”, while however “to remain proportionate, it is appropriate to envisage that this marking obligation should not cover AI systems performing primarily an assistive function for standard editing or AI systems not substantially altering the input data provided by the deployer or the semantics thereof”.
Content detection mechanisms (CDM), marking content as AI‑assisted for the benefit of end‑users’ awareness, should therefore be embedded in AI systems, according to a technology‑neutral approach, which refers to the ever‑evolving state of the art as the benchmark against which the appropriateness of such CDM should be assessed.
In this vein, Recital 134, in addition to the technical solutions to be adopted by providers, requires that deployers using AI systems “to generate or manipulate image, audio or video content that appreciably resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (deep fakes), should also clearly and distinguishably disclose that the content has been artificially created or manipulated by labelling the AI output accordingly and disclosing its artificial origin”.
Of particular interest is that the same Recital 134 places this transparency obligation in relation to the right to freedom of expression, as a fundamental right enshrined in the constitutions of Member States, as well as in the Charter of Fundamental Rights of the European Union. Recital 134 clarifies that the obligation to disclose the use of AI systems in the creation of deep fakes should not stand in the way of freedom of expression and freedom of the arts and sciences, “in particular where the content is part of an evidently creative, satirical, artistic, fictional or analogous work or programme... In those cases, the transparency obligation for deep fakes set out in this Regulation is limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work, including its normal exploitation and use, while maintaining the utility and quality of the work”.
This should be read as meaning that, to the extent that freedom of the arts, science and expression are at stake, the transparency obligation should be applied reasonably, so as not to disrupt or deprive of effectiveness the output concerned, particularly when it is the expression of artistic endeavour. One may think, for example, of the usage of deep fakes in movies: it is quite obvious that obliging film‑makers to conspicuously display on screen, while the audience is watching a scene, that certain realistic elements are AI‑generated deep fakes would alter the underlying creative intention and unreasonably interfere with the enjoyment of a work of art which the audience will from the outset perceive as purely fictitious and not informative. To think otherwise, we would already have conspicuous on‑screen warnings about the non‑existence of Jabba the Hut and all the other bizarre creatures populating Chalmun’s Spaceport Cantina in the 1977 Star Wars Episode IV: A New Hope.
Finally, Recital 134 imposes similar disclosure obligations “in relation to AI‑generated or manipulated text to the extent it is published with the purpose of informing the public on matters of public interest unless the AI‑generated content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for the publication of the content”.
The principles set out in the above recitals are then framed as proper normative precepts in Article 50 of the AI Act, whose paragraphs restate those principles.
On 17 December 2025, the EU Commission published the first draft of the Code of Practice on Marking and Labelling of AI‑generated Content, in pursuance of paragraph 7 of Article 50 of the AI Act – the Code can be accessed here. The draft Code provides for certain technical principles to be followed by stakeholders in advancing the marking of AI‑generated content, in terms of, in summary, machine‑readable marking techniques, non‑removal of machine‑readable marking, and the effectiveness, reliability, robustness and interoperability of marking and detection techniques.
Returning to the impact that the concealed use of AI systems for content creation and manipulation can have on fundamental rights, beyond the right to fair elections and access to information, AI evolution could also undermine other fundamental human rights, such as the right to freedom of thought. This right encompasses the right to keep our thoughts private so that we may not be coerced into revealing them, freedom from manipulation, and a prohibition on penalising persons for their thoughts or opinions alone.
The ability to manipulate is ingrained in these technologies and makes them potentially very offensive to the right to freedom of thought. If an individual is manipulated through targeted fake content, this is likely to result in instances of breach of this fundamental right. It is unlikely that fake information or content is passed off as genuine without also being used to manipulate target individuals. At a minimum, such manipulation consists in deception, whose gravity should be measured depending on the relevance and impact of the distorted message resulting from AI manipulation (for example, a deep‑fake manipulation concerning a political candidate could lead to the election of the opponent, supported by a foreign state entity interested in advancing its imperialist stance and gaining political support in the country where elections take place).
By way of conclusion, it is worth mentioning certain research findings on human interpretations of synthetic content, reported in a 2023 OECD paper on Initial Policy Considerations for Generative AI. According to these findings, humans trust content less when AI authorship is disclosed, while they find synthetic faces (paradoxically) more trustworthy than real faces. This has been explained by the fact that synthetic images resemble “average” faces, which are perceived as more trustworthy.
These findings per se – even leaving aside all the preceding remarks – are sufficient to support the need for sound transparency obligations on AI‑generated output. The AI Act’s legislative solutions seem sufficiently balanced and forward‑looking, reflecting a generalised consensus on the undesirable and serious interference that the undisclosed use of AI may have with fundamental human rights.
Before looking into what AI system providers and/or deployers are by law required to do when they use AI to generate or manipulate content, the cognitive and societal context underlying the legal issue should be examined. The initial question is why users would need at all to know they are interacting with AI or that a given content was AI generated or manipulated. The answer to this question implicates fundamental issues of democratic participation and individual dignity. Trust, authenticity, user empowerment, responsibility and agency are at the core of the legal and ethical discourse, as the values that may be at stake when the use of AI systems for content production is concealed.
Truth, Trust and AI: The Case for Transparency in AI-Generated Content
Written by Massimo Maggiore
emlex – Eva Maschietto Massimo Maggiore
FEATURED ARTICLE
ITALY
Massimo Maggiore is co-founding partner, together with Eva, of emlex, of which he heads the intellectual property/TMT and competition/commercial practices department. He is academic fellow of Milan’s Bocconi University as contract professor at the course Cyber risk and data protection law, lecturing on the cybersecurity related legal framework. Massimo also lectures at the same University’s LL.M in law of internet technology on cloud computing related laws and also regularly serves as professor at the Milano Fashion Institute’s master on fashion law and other courses. He has many years of experience in advising fast moving consumer goods (FMCG) industries with a focus on business practices, competition and related legal issues such as marketing, advertising and brand protection.
Massimo has engaged the in-depth study of legal issues relevant to e-commerce, such as the conclusion of online contracts, cybersecurity, vertical competition and the Digital Markets Act, the protection of personal data with specific regard to processing in the context of online trading activities. He is the author of several publications and a frequent speaker at conferences in Italy and abroad on intellectual property, information technology and unfair commercial practices in the food supply chain.
.
Biography
GLOBAL
RECOGNITION