Can input secretly view others' chat records? The "vulnerability" in DeepSeek has been seriously misunderstood.
According to Beating reports, there are rumors that inputting specific tags into DeepSeek allows users to see others' conversations, which has been characterized as a failure of P0 multi-tenant isolation. In fact, this is a phenomenon where the model generates seemingly real conversations based on memory and system prompts, rather than real-time retrieval of others' sessions. It is a common issue related to training data memory extraction and is not unique to DeepSeek. Academic research has indicated that training data can be extracted through template tokens; the presence of the current date in conversations makes it difficult to prove they originate from others. Currently, to confirm actual user data leakage, concrete evidence is needed. No evidence currently points to the existence of other users' information.

