Even with “Memory” and style preferences turned off in the ChatGPT personalization settings, the paid “Team” subscription tier causes personal data to be disclosed to the underlaying Language Model. This includes:
- Full name used for registration
- Corporate E-Mail address
- “Handle”: corporate user name
- Organization name
(a November 2024 report by Prof. Florian Gallwitz showed that the LLM was also passed the IP address)
This background information can induce certain biases, scew or even ellicit confabulation/hallucination as shown here: when inquiring about an old patent filing of mine, ChatGPT claimed that the technology from the patent, developed at my former employer, was used in the tech stack of my new employer (something I don’t know but highly doubt). I suppose the connection to my new employer was made based on the Organization Name as disclosed via the model context.
The free version of ChatGPT does not seem to have this problem:
I wonder what it is like elsewhere - ChatGPT Enterprise, Claude, …
Written on July 20th, 2025 by Nils Durner. Last updated: July 20th, 2025