Russian Police Warn ChatGPT Builds Digital Dossiers on Users

ChatGPT can store personal details in its personalized memory, effectively assembling a digital dossier on users, according to the Russian Interior Ministry's unit combating criminal use of information technology.

Citing a study titled The Algorithmic Self-Portrait, the ministry said specialists found that OpenAI saved users' personal information and built digital portraits of them despite its own data privacy policy. Nearly all such records were created without the person's consent.

Most accounts had names, addresses, financial details and social status held in the AI's memory. Half of users had information about their health and beliefs stored, while almost every account had the AI remembering goals, intentions, preferences and behavioral traits.

The ministry said this accumulation creates potential risks of identifying users even when they use pseudonyms. Combined data on professional activity, health, income and personal experiences could be used to build a digital dossier.

Another threat is manipulation of AI memory: false instructions inserted into processed documents could let the AI covertly influence officials and scientists, factoring in their personal data and psychological traits. The main danger is that such influence is invisible and targeted at a specific person.

To protect their data, the ministry advises checking memory and personalization settings in AI services, avoiding uploads of work documents and personal information, anonymizing queries and using temporary chats.