Liechtenstein DPA warns ChatGPT ads may expose users’ intimate information
OpenAI’s planned Europe rollout would use chats, usage and device data for opted-in ads, while generic advertising could still use current chat context.
The Data Protection Authority of Liechtenstein issued a notice warning about OpenAI’s planned personalised advertising in Europe. OpenAI intends to offer the feature in both the free version of ChatGPT and the paid ChatGPT Go subscription, but only after users give opt-in consent. According to the authority’s account of OpenAI’s information, personalisation could use past and current chats, usage data and device information.
The notice matters because a chat can reveal substantially more than ordinary browsing or search behaviour. A single travel-planning conversation could disclose a destination, budget, family situation, health needs and personal preferences. The authority says such content may allow conclusions about a person’s intimate sphere or current mood. For businesses that permit ChatGPT use, the supervisory signal is that chat history is a high-context source of personal information, not merely another behavioural advertising signal.
Two advertising models
The authority distinguishes between personalised and generic advertising.
Users who do not consent to personalisation would continue to receive generic advertisements. Those ads may nevertheless be based on the current chat and limited context such as location and time of day. According to the notice, OpenAI apparently would not use an existing user profile, or create a new one, for that purpose.
OpenAI relies on an overriding legitimate interest under Article 6(1)(f) of the GDPR for this more limited advertising processing, according to the notice. Personalised advertising, by contrast, is described as requiring consent. Declining the personalised advertising option therefore would not necessarily mean that the current conversation has no relevance to advertising.
The Liechtenstein authority says its explanation draws on OpenAI’s EU privacy policy and FAQ, together with assessments and recommendations from the data protection authorities of Berlin, Lower Saxony and Hamburg. Its wording is advisory: it recommends privacy-protective steps rather than announcing a penalty, final rule or new compliance deadline.
Atlas reads the notice as identifying the informational richness of AI conversations as the central regulatory concern. The issue is not only whether a consent button is presented, but whether users understand what personal circumstances their chats may disclose before they provide the data.
Settings the authority recommends
For users seeking to limit advertising personalisation and use ChatGPT more sparingly, the authority recommends:
- refusing consent to personalised advertising;
- checking the privacy settings and disabling advertising personalisation if it is active;
- disabling the “Memories” function and deleting existing memories;
- disabling the use of chats for model improvement;
- using “Temporary Chat” for sensitive subjects;
- not entering confidential or secret information;
- deleting chat history regularly; and
- using an ad-free version of ChatGPT.
The recommendations extend beyond advertising settings. The authority says privacy protection begins with the decision about what information to disclose to an AI system, rather than only with the later choice of whether that information may be used for personalised ads. For compliance teams, that makes user guidance, internal acceptable-use rules and controls around confidential information relevant even where personalised advertising is not enabled.
The notice’s most concrete warning is that a conversation can expose a person’s intentions, circumstances and sensitivities in one record. That risk exists at the point of disclosure, before any advertising preference is selected.
Permanent link to this Atlas analysis
Related Atlas analysis
Continue with coverage connected by market, topic and operating context.
- European UnionEuropean Parliament calls for AI oversight of EU democratic processesMEPs voted 405 to 133, with 60 abstentions, and proposed a budget-neutral AI observatory for parliamentary scrutiny.
- EstoniaEstonian government sends AI agent code draft to parliamentThe proposed identifier would link an agent to a specific person without granting rights or automatic access to data or services.
- United StatesCFTC sues Louisiana over order affecting more than 30 companiesThe Commission says the Thursday order came without notice or opportunity to be heard, and seeks to preserve federally regulated trading.
Continue with Atlas
Move from this development into the relevant research, comparison and workflow.