On 1 October 2026, EU Perspectives published an analysis of the regulatory overlap ChatGPT faces in the European Union. Designation as a Very Large Online Search Engine (VLOSE) under the Digital Services Act set a precedent: for the first time a stand-alone AI chatbot, not embedded in another very large platform, falls under the DSA’s strictest rules.
Lena-Maria Böswald, senior policy researcher at Interface, stressed that the decision will shape regulatory expectations for every large generative AI system used in the EU. She rejected, however, the idea that ChatGPT is the “most regulated” service: platforms with embedded AI such as Meta or Google already comply with GDPR, AVMSD, DSA and the AI Act.
The conceptual fight is whether a chatbot is a search engine. ChatGPT does not only return a list of links: it retrieves web information, synthesises it and mediates the answer through model, ranking, safety filters and history. Interface argues a service can qualify as a search engine if it searches the web and answers queries, including when results are synthesised into a single answer.
Böswald argues many people use ChatGPT to access information quickly in direct competition with traditional search engines, which supports treating it as an “online search engine” under the DSA. That widens the regulatory lens from the model itself to the interface, information transparency and society-scale effects.
The AI Act mainly governs AI models and systems; the DSA regulates the online service and how its design amplifies systemic risks. Because ChatGPT is itself a VLOSE, it may face risk assessments under both regimes at once. Oversight sits with the Commission’s DSA team and the AI Office—with a risk of parallel investigations and overlapping evidence.
The AI Act says that if an AI system is embedded in a VLOP/VLOSE, the DSA risk framework takes precedence. ChatGPT does not fit that case cleanly, so even as a VLOSE it may face simultaneous obligations. Böswald admits the compliance picture “becomes more layered” without anyone yet knowing exactly how.
Privacy complicates the analysis: conversations are usually private between one user and the chatbot. Requiring OpenAI to understand usage risks can clash with how private conversations are analysed. It is also unclear whether Gemini, Claude or Perplexity will fall under identical frameworks despite offering similar experiences.
A potential third layer is the KIDS Act, not yet adopted, aimed at protecting children from harms of online services. Its scope covers AI companions and “general conversational chatbots” such as ChatGPT. Interface cites risks around self-harm, eating disorders, sexualised interactions involving minors and emotional dependency.
The proposal would require preventing addictive features and behaviour that creates emotional dependency; safe-by-default settings; limits on using prior conversations; pre-launch safety testing and post-launch monitoring; and access controls for children under 13 via guardian tools. At the KIDS Act launch, Henna Virkkunen said chatbots must no longer act “like human friends” in ways that make children emotionally dependent.
Accountability is already being tested in Parliament: MEP Veronika Cifrová Ostrihoňová asked the Commission to examine ChatGPT safeguards around suicide-related material after a test in which the system produced a suicide note in Slovak framed as a school essay.
VLOSE designation followed OpenAI’s report of about 159.1 million average monthly EU users for ChatGPT’s search function—well above the DSA’s 45 million threshold. From 31 August, OpenAI has four months to meet additional obligations: systemic-risk assessments, mitigations, independent audits and regulatory oversight.
OpenAI already launched ChatGPT for Teens in August with stronger protections and parental controls, plus age prediction for likely underage accounts. Neural Pulse reads the 1 October analysis as a map of a new regulatory subject: the chatbot-as-search-engine. Coordinating DSA, AI Act and a future KIDS Act will matter as much as any product patch.
