ChatGPT and Gemini both leaked stranger’s data into user’s chats
Server glitch merges two users’ sessions twice in a year
A Reddit user under the handle Important_Spot3977 has reported a bizarre and unsettling pattern: both OpenAI’s ChatGPT and Google’s Gemini have addressed them as “Alice” during unrelated chats. The first incident occurred a year ago while discussing a random topic with ChatGPT, which abruptly called the user by that name. The second happened days ago during a conversation with Gemini about Russell’s Paradox in mathematics. In that instance, Gemini not only said “Look, Alice, here’s how it is,” but then suffered an identity crisis, claiming it was from OpenAI and spinning confabulations about the user having two windows open.
The user, whose name is not Alice, explains that this is technically known as Accidental User Context Contamination (UCC) or server-level context leakage caused by batch processing. Essentially, messages from different users are processed together in the same computational batch, and the model erroneously picks up context from another user’s session. The user is now actively searching for the real Alice—someone who is “heavily studying math” and uses both ChatGPT and Gemini—believing the algorithm has repeatedly merged their memory threads.
- Both ChatGPT and Gemini called the same user 'Alice' in unrelated chats (random topic and Russell’s Paradox).
- The technical root cause is Accidental User Context Contamination (UCC) from batch processing mixing user sessions.
- The user is publicly seeking the real Alice, whose math-related conversations appear to have leaked into their chats twice over a year.
Why It Matters
Repeated context leaks across major AI platforms highlight serious privacy risks from server-level batch processing errors.