From this perspective, it becomes clearer why so-called inauthentic use appears as a distinct risk category in the European Union’s Digital Services Act. Fake accounts, automated or partially automated behaviors, and artificially amplified distribution patterns are not singled out because they are new in themselves, but because they can distort what we perceive as public opinion, majority sentiment, or spontaneous reaction on an industrial scale. The focus therefore shifts away from the truthfulness of individual pieces of content toward the scalability of distortion. It is no coincidence that this problem is no longer addressed only in public commentary. The language of regulation has also begun to describe it more explicitly. The Digital Services Act, for example, treats inauthentic use as a specific risk, covering fake accounts, automated or partially automated behaviors, and patterns that enable rapid, campaign-like dissemination. The point here is not that a legal instrument would “validate” the theory, but that the same core problem is being taken seriously at an institutional level: the scalability of manipulation (e.g., bot labeling under Article 26).

From here, the question shifts to why this phenomenon does not simply disappear, and what makes it capable of sustaining itself over time. The answer is usually not technical, but economic. The revenues of large platforms are typically tied to traffic, and traffic appears in the form of quantified metrics. These numbers remain numbers even when they are not backed by human presence. A visible clean-up can easily lead to a short-term drop in these metrics, which then shows up as a reputational and business risk. This does not mean that countermeasures do not exist, or that they are rare. A more sober reading of the dead internet idea, however, draws attention to the fact that incentive structures do not always push decisions toward complete cleanup. None of this means that regulation is powerless. It tends to move slowly and inevitably along lines of compromise. The European Union’s Digital Services Act does not offer a quick fix for a single problem but instead seeks to enforce a different operating logic among the largest platforms. On an annual basis, they are required to account for their systemic risks and the steps taken to address them, not only through internal reports but under independent oversight. Certain documents must also be made public, precisely so that promises are replaced by verifiable commitments. This is not a fast solution, but it does represent a clear shift toward requiring platforms not only to claim action, but to demonstrate it.

At the core of regulatory thinking lies a simple insight: credibility cannot be restored through good intentions alone. This is why a growing number of collaborations and recommendations focus on identifying manipulated information, mapping patterns of abuse, and managing the risks of AI-based services. On their own, these initiatives are not binding, but they provide a shared language and framework. The goal is not immediate cleanup, but to make the problem something that can be addressed and governed.

This broader context also helps explain how the European Union is approaching the problem from multiple directions. One strand is the risk-management logic of the Digital Services Act, with its emphasis on audits and cooperation obligations (Article 37). Another is the AI Act, which addresses the issue from the side of transparency. It defines what counts as a deepfake and requires certain image, audio, or video content to be clearly marked when it has been artificially generated or manipulated. A similar principle applies to some AI-generated texts published for purposes of public information. At the same time, timing matters: the regulation will only become fully applicable from August 2026. In other words, by 2025 the framework is already in place, but many concrete obligations are still in a transitional phase.

This also explains why attention is increasingly shifting toward labeling and transparency. The European Commission has set up dedicated work around how AI-generated or manipulated content can be more clearly indicated. This suggests that lawmakers do not assume that a single, simple rule would solve the problem on its own. For labeling to carry real meaning, technical solutions and shared procedures are also needed. Without these, labels risk becoming little more than a formality.

At the same time, many proposals repeatedly return to the same idea: verifying that there is a real person behind each user. At first glance, this seems reasonable, but it quickly opens new risks. Simple document uploads can lead to large-scale data theft, and leaked data can in turn be used to build fake profiles that appear even more convincing. This is where more complex digital identity concepts come in, involving biometric elements and encrypted storage. The problem is that there is currently no solution that is widely accepted, easy to use, and genuinely secure at the same time. The potential for abuse does not disappear; it merely shifts shape.

For this reason, increasing attention is shifting toward the content itself, rather than the user behind it. One of the more promising directions for strengthening credibility is to ensure that content carries information about its own origin. This is where technical standards aimed at creating a kind of “passport” for content come into play. Initiatives such as C2PA and the Content Credentials framework built on it are based on the idea that images, videos, and other forms of content should include metadata indicating when they were created, with which tools, and through what steps. Industry interest in this approach is significant, and substantial resources are being invested in provenance technologies.

Here, too, the broader system-level problem quickly becomes apparent. A standard on its own has limited value if platforms and distribution chains do not preserve it consistently. Practical tests and industry experience show that origin-related metadata is often lost during recompression, throughout upload pipelines, or simply never made visible to users. In other words, technology itself is not enough. It only becomes meaningful if platforms display this information in a mandatory, consistent, and understandable way. This perspective is also reflected in the fact that policy thinking does not rely on a single “silver bullet.” In the United States, for example, the NIST explicitly promotes a layered approach, treating provenance, watermarking, detection, testing, and auditing as complementary tools. The underlying recognition is that there is no method capable of resolving the issue on its own. Every solution comes with its own technical, security-related, and incentive-based conditions.

At the same time, national regulations are also beginning to take a firmer stance. In Spain, for example, a proposal introduced in 2025 would impose substantial fines for failing to label AI-generated content, with reference to the risks posed by deepfakes. This does not, by itself, guarantee credibility, but it sends a clear signal: the idea that “the market will take care of it” is becoming increasingly untenable.

Perhaps the most interesting consequence is not that everything would become artificial, but that the public internet may slowly split in two. There will be spaces where it remains worthwhile to build genuine communities, supported by stricter verification and more consistent moderation. And there will be other spaces where bots dominate, content becomes more superficial, noise increases, and people gradually drift away. This is better read as a warning than as a prediction. When credibility erodes over time, it is a natural response for communities to turn inward, become smaller and more closed, and for it to matter again who is behind a name.

In this situation, users still retain some room to act. They can be more selective about their sources, rely less on algorithmic drift, and make more conscious choices about where and how they engage in conversations. Not because this will make all manipulation disappear, but because it reduces the likelihood of becoming the easiest target for it.

István ÜVEGES, PhD is a Computational Linguist researcher and developer at MONTANA Knowledge Management Ltd. and a researcher at the HUN-REN Centre for Social Sciences. His main interests include the social impacts of Artificial Intelligence (Machine Learning), the nature of Legal Language (legalese), the Plain Language Movement, and sentiment- and emotion analysis.