German Academic Loses 2 Years of Research Notes After ChatGPT Privacy Toggle
Updated
Updated · TechBullion · May 22
German Academic Loses 2 Years of Research Notes After ChatGPT Privacy Toggle
1 articles · Updated · TechBullion · May 22
Marcel Bucher said two years of research notes, draft papers and funding applications vanished after he disabled ChatGPT’s setting allowing chats to be used for model training in August 2025.
OpenAI support told him its “Privacy by Design” approach limited data access and recovery, leaving the missing conversation history irretrievable once it disappeared.
The exact cause remains unclear: independent checks suggest turning off training does not automatically delete earlier chats, raising the possibility of a bug, account issue or separate deletion event.
In Nature, Bucher warned academics against treating ChatGPT as a primary workspace, arguing AI chat tools should be backed up and used as companions rather than document archives.
Beyond data loss, what deceptive behaviors and hidden security flaws are AI models now known to conceal from their users?
AI tools can void legal privileges and leak secrets. Are professionals sleepwalking into a compliance and confidentiality nightmare?
When Privacy Deletes Progress: The 2025 ChatGPT Data Loss That Cost a Professor Two Years of Work
Overview
In August 2025, Professor Marcel Bucher suffered a major data loss when he disabled the data consent option in ChatGPT, leading to the irreversible deletion of two years of his academic work. This incident, later detailed in Nature, highlights the hidden risks of generative AI tools and the critical need for strong data management practices. It shows that privacy settings can unexpectedly erase valuable work, and stresses the importance of regular backups and careful review of privacy options. Professor Bucher’s experience is a warning for all users to balance innovation with robust data protection.