Multiresistent, extrem überlebensfähig – das Bakterium Acinetobacter baumannii bereitet Ärzten weltweit Sorgen. Ausbrüche wären vermeidbar. Denn erst die Medizin hat den Keim zum Problem gemacht.

Genmanipulation hat das Interesse der Tech-Milliardäre geweckt. Sie finanzieren Firmen, die Eingriffe in die DNA von Embryonen planen. Damit könnten sie nicht nur ein Leben verändern, sondern die Zukunft vieler Generationen.

Warum wird der eine süchtig und der andere nicht? Manche Menschen hätten eben einen Suchtcharakter, hiess es früher oft. Doch was Menschen wirklich in die Abhängigkeit abrutschen lässt, ist wesentlich komplexer.

Störche spüren Mäuse tief in der Erde auf und holen sie mit einem gezielten Zustechen ihres Schnabels hervor. Wie sie das schaffen, ist ein Geheimnis. Die Kolumne «Wild & wundersam».

Die Geopolitik hat die arktische Insel ins Zentrum der Aufmerksamkeit gerückt. Für die Wissenschaft ist Grönland seit langem wertvoll, denn unter dem Eisschild liegt ein ganz besonders Klima- und Erdarchiv. Daraus lässt sich zum Beispiel ablesen, wie empfindlich der Eisschild einst auf Wärme reagiert hat.

In October 2025, following OpenAI’s disclosure that ChatGPT’s search feature had reached an average of 120.4 million monthly users in the EU, a Commission spokesperson confirmed that regulators are currently assessing whether ChatGPT can be designated as a Very Large Online Search Engine (VLOSE) under the Digital Services Act (DSA). With nearly three times the active users as the statutory threshold, the quantitative case is unambiguous. The legal question is whether a service that synthesises answers rather than returning indexed links falls under Article 3(j) DSA as an “online search engine”. The Commission should answer yes. A functional interpretation is legally mandated by the DSA’s risk-based architecture, economically justified by generative AI’s displacement of traditional search, and urgently necessary given the enforcement vacuum in transatlantic AI governance. Excluding ChatGPT from VLOSE designation would privilege form over function and leave the EU’s most significant AI-mediated information risks unaddressed.

Why chatbots classify as VLOSE legally

The DSA is a comprehensive set of rules for online services provided within the internal market, which prioritises the safeguarding of fundamental rights and consumer protection. Where an online service is designated as a VLOSE under the DSA, it must adhere to certain obligations in order to ensure a safer and more predictable market for users within the EU. Article 3(j) of the DSA defines a VLOSE as an “intermediary service that allows users to input queries” for search purposes across all links and websites and accordingly “returns results in any format in which information related to the requested content can be found”. With the advancement of AI and the rapid rise in the use of online chatbots such as ChatGPT and Gemini, the question arises as to whether these AI-driven services should be subject to the high level of compliance mandated by the DSA.

The threshold for VLOSE is 45 million users on average per month (Article 33(1)), a figure that ChatGPT has far surpassed, reporting an average of 120 million EU users per month. Where this jurisdictional threshold is met, there is a strong presumption that such services fall within the scope of the DSA, if there is a “substantial connection to the Union”.

In order to determine whether its functionalities align with the legal definition set out in Article 3(g)(iii) of the DSA, one must determine whether ChatGPT is an intermediary service. Concretely, whether it falls under the category of “hosting service”, which means that it stores information provided by and at the request of a recipient. This is the case when the search function of ChatGPT is separated from its other features. Traditional search engines present a list of results ranked in order of what best aligns with key phrases within a given query. AI chatbots go beyond key phrases, incorporating the most up to date information, as well as stored feedback, in order to enhance results which are personal to the user. This storage of feedback should be interpreted as falling under the category of a hosting service.

While ChatGPT does allow users to input queries to perform searches, rather than provide links from all websites, these chatbots essentially provide a shortcut for users by preemptively scanning relevant websites to provide an auto generated response to the user. Additionally, chatbots directly give sources if asked, so they are functionally similar to search engines. Article 3(j) of the DSA however does not specify that a VLOSE must provide links and websites, but rather that it returns results in any format relevant to a given query. This is important to note as more and more generative AI systems are being embedded into platforms which are already subject to the DSA, a point which the Commission has noted for consideration. The complex nature of given responses from such chatbots therefore aligns with this aspect of the legal definition of a VLOSE.

Since the enactment of the DSA, the Commission has designated Google Search and Microsoft Bing as VLOSEs. In both cases it was found that the jurisdictional threshold was exceeded and both were identified as intermediary services under Article 3(g), as well as search engines under Article 3(j). ⁠No court has yet interpreted Article 3(j) in the context of generative AI. However, the Commission’s approach in recent proceedings suggests openness to functional interpretation. In its December 2025 decision fining X €120 million under the DSA, the Commission applied a “functional approach“ to the definition of “provider”, looking to substance over formal classification (para 47). A similar interpretive method would support VLOSE designation for services that perform search functions regardless of their technical architecture.

Furthermore, because of the advanced nature of these chatbots, one cannot ignore their role in retrieving, processing, and disseminating information. The DSA addresses “systemic risks” including threats to fundamental rights, public safety, democratic discourse, and societal well-being. Increased transparency, due diligence obligations and risk mitigations could help to address the many concerns surrounding the potential impact of chatbots on “emotionally vulnerable individuals” as well as biases integrated into chatbots which lead to harmful outputs.

The benefits of classifying chatbots as VLOSE

Classifying AI chatbots as VLOSE creates a new economic reality. This designation addresses critical market failures that threaten the information system. It curbs harmful advertising models and creates a marketable trust premium for compliant firms. The economic case is compelling, though not without trade-offs. The strongest argument concerns the “Zero-Click” crisis. AI platforms extract value from publishers’ content while externalising the cost of content creation. Users increasingly rely on AI-generated summaries rather than clicking through to sources. The scale of this displacement is measurable. The Reuters Institute finds publishers expect search traffic to decline by 43% over the next three years, compounding a trend that has already seen traffic from Google fall by 33% globally. The threat from chatbots is acute: according to research from TollBit cited by the News Media Alliance, their click-through rates are 95.7% lower than traditional search.

VLOSE designation addresses this through Article 34(1)(d) of the DSA, which requires platforms to assess systemic risks to media pluralism and civic discourse. Combined with the mitigation obligations under Article 35 of the DSA, this compels platforms to evaluate how their design choices (summarisation, content extraction, traffic diversion) affect information diversity. This is not a direct compensation mechanism; that role falls under Article 15 of the Copyright Directive. But it creates regulatory pressure to internalise costs that platforms currently externalise. Designation also pushes platforms away from harmful advertising models. Systemic risk framework under Article 34 of the DSA captures manipulative advertising design as a risk to be assessed and mitigated.

Research from MIT, co-authored by 2024 Nobel laureate Daron Acemoglu, finds that the welfare benefits of ad-based models versus subscription models are conditional on how intrusive users perceive the ads to be. After switching from surveillance-based to contextual advertising, the Dutch public broadcaster NPO saw revenue increase by 79-149%. The DSA does not ban surveillance advertising outright, but designation creates structural incentives toward less extractive models. Furthermore, compliance creates a degree of marketable “trust premium”, with the World Economic Forum confirming that investors reward strong AI governance. More broadly, the VLOSE framework is designed to have systemic effects: correcting market failures, protecting content creators from AI-driven value extraction, and eliminating harmful advertising models. Proponents argue that far from being a mere compliance burden, the classification positions the EU as a laboratory for sustainable digital market structures that may become global standards.

While these considerations appear economic or political, they are directly relevant to the legal analysis. The DSA is a risk-based framework, where designation is contingent on identifying “systemic risks” to the public interest (Article 34). The economic displacement of traditional media, the market-distorting effects of certain advertising models, and the high costs that may entrench incumbent power are all factors in assessing whether a service poses a systemic risk to media pluralism, consumer welfare, and market fairness. The economic data, therefore, is not merely context; it is the evidentiary basis for a functional and purposive interpretation of the law.

Risks and costs

As the EU considers VLOSE designation for ChatGPT, required Risk Assessments create a market for trustworthy AI. Compliant firms are well placed to benefit from the growing digital trust market, projected to reach $130-470 billion by 2035. On the contrary, the designation carries major economic risks. Compliance costs are substantial. Estimates range from $150 million by CCIA Research Center to €260 million annually by Bertelsmann Stiftung for a single large platform. The designation also risks entrenching incumbents. While imperfect as an analogy, GDPR applied universally rather than asymmetrically, its empirical record offers a cautionary signal. After GDPR’s implementation, market concentration increased by 17%. An Oxford Martin School study found small IT firms saw profits drop by 12.1%, far more than the 4.6% drop for large firms. The mechanism differs under a threshold-based designation, which targets incumbents rather than burdening the entire market, though the risk that compliance costs deter new entrants remains real. The Draghi Report noted that 30% of European unicorns have relocated abroad, driven in part by regulatory load. Research from NBER found GDPR caused a 20.63% reduction in EU deals led by US investors, representing over $1.58 billion in lost capital annually. These risks must be weighed against the benefits of a healthier, more transparent digital market, but they cannot be dismissed.

The (geo)political context makes DSA enforcement essential

The EU’s decision to classify and enforce rules on US chatbots is a calculated geopolitical risk. The US has reacted aggressively, framing DSA and DMA enforcement as discriminatory. It is preparing a Section 301 investigation, the same tool used for tariffs on China, and has imposed visa bans on EU officials involved in tech regulation, including Thierry Breton. Despite these threats, the EU maintains that regulatory sovereignty is not negotiable. Competition Commissioner Teresa Ribera has made this position explicit, stating that the bloc will not reverse its regulatory framework under external pressure and must be prepared to walk away from trade negotiations. This stance is bolstered by game theory analysis from Bruegel, which concludes that since any fine could be a pretext for US retaliation, lowering fines is strategically pointless and only harms regulatory credibility without reducing the risk of escalation. The EU has demonstrated this resolve in practice. Following US pressure over Greenland, the bloc walked away from the Turnberry Agreement, a proposed transatlantic trade and security negotiated in early 2025. That decision signalled that regulatory sovereignty would not be offered as a concession in broader diplomatic bargaining. The EU is not without leverage. It has a prepared €93 billion tariff package and a history of strategically targeting US goods in trade disputes. More fundamentally, the EU’s own enforcement record constitutes a massive countervailing force. In 2024, the Commission levied approximately €6.7 billion in combined competition, DMA, and DSA fines against US technology firms. This figure aggregates penalties across multiple enforcement instruments and authorities. The potential DSA penalties amplify this structural leverage further. Under Article 74 DSA, fines can reach up to 6% of a company’s global turnover, meaning a firm like Google faces theoretical exposure in the tens of billions. This is not leverage the EU needs to threaten; it is leverage that exists as a permanent feature of the regulatory landscape. Enforcement authority for VLOP under the DSA rests with the European Commission itself, specifically DG CONNECT.

The institutional design creates a structural ambiguity. On one hand, political control enables geopolitical calibration of enforcement intensity, indicating the Commission can modulate procedural pace and fine severity in light of broader diplomatic considerations. On the other hand, this same structure exposes enforcement to the very geopolitical pressure the US is applying. As Verfassungsblog has noted, the absence of an independent enforcement agency raises legitimate questions about whether proceedings can be fully shielded from political influence. This tension is the central vulnerability in the EU’s otherwise strong position. Crucially, however, DSA enforcement cannot function as a rapid-response instrument. Investigations typically take 18–24 months, based on observed proceedings, and are structured around procedural requirements, including “preliminary findings” under Article 73(2) and the respondent’s “right to be heard”. The legal classification of chatbots provides a constant, structural countervailing force against US trade pressure, but it is slow, steady pressure, not a tactical weapon that can be deployed in response to new tariffs.

Legally sound, economically justified, and geopolitically coherent

The case for designating ChatGPT as a VLOSE is legally sound, economically justified, and geopolitically coherent. Functional definition under Article 3(j) of the DSA encompasses services that retrieve and synthesise web information at scale, regardless of whether they return traditional link lists. The economic benefits – such as transparency requirements, systemic risk assessments, and pressure toward sustainable advertising models – address genuine market failures. And the political risks, while real, do not counsel retreat. Game-theoretic logic and recent practice confirm that regulatory credibility depends on consistent enforcement. What remains underdeveloped is the enforcement architecture itself. The Commission’s dual role as a political body and sole enforcer creates structural vulnerability. The EU should consider whether independent oversight mechanisms could insulate VLOSE proceedings from the geopolitical pressures that will inevitably intensify as designation decisions multiply.

The post Searching for Answers appeared first on Verfassungsblog.