Skip to content
Submit incident
AII-1001-00239
Well documentedThree or more independent publishers report the core facts.Supporting sources are pending verification.

Professor loses two years of academic work after disabling ChatGPT data consent

A plant sciences professor at the University of Cologne reported that two years of academic work permanently disappeared from ChatGPT Plus after he disabled the platform's data consent setting in August. He had used the tool daily for grant applications, publication revisions, lectures, and other academic tasks, but was unable to recover the data despite attempting recovery steps and contacting OpenAI support.

August 2025 Β· OpenAI Β· Reliability and hallucination
August 2025
AII-1001-00239
Quick facts
Category
Reliability and hallucination
Country
Germany
Date
August 2025
Drafted by
Raidu Incident Detection

What happened

Marcel Bucher, a plant sciences professor at the University of Cologne, reported that he lost two years of academic work after disabling the 'data consent' option in ChatGPT Plus in August. When he disabled the feature to test whether he would retain access to functions without providing data to OpenAI, all of his chats and project folders were immediately and permanently deleted with no warning or undo option.

Reported impact

Bucher reported losing carefully structured academic work spanning two years, including grant applications, publication revisions, lectures, exams, and email drafts. He had saved only partial copies of some conversations and materials, with large parts lost permanently.

Index analysis

Bucher wrote that he valued ChatGPT's ability to remember conversation context and the stability of the workspace. He relied on it daily as an assistant but acknowledged that large language models could be inaccurate and did not equate the tool's reliability with factual accuracy.

Governance controls

Bucher attempted to recover his data by reinstalling the app, trying different browsers, and adjusting settings further. He contacted OpenAI support but was initially routed to an AI agent that could not help, and a subsequent human representative was also unable to assist in data recovery.

Classification

Category
Reliability and hallucination
Industry
Education and training
Country
Germany
Harm status
Actual harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Economic or property
Harmed
Workers
Consequences
Content removal
AI systems
ChatGPT Plus

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident