GPT-2 Reportedly Reproduced Personal Data from Its Training Data

February 14, 2019

OpenAI's GPT-2 reportedly memorized and reproduced portions of its training data, including personal information such as names, email addresses, social media handles, and phone numbers. Researchers raised concerns that large language models could expose private or sensitive information when trained on web-scale datasets containing personal data.

This incident highlights the need for stricter data governance and AI governance frameworks to prevent such incidents in the future. To learn more about responsible AI practices and project Cerebellum, visit JOIN US.

Matched TAIM controls

Suggested mapping from embedding similarity (not a formal assessment). Browse all TAIM controls

Alleged deployer
OpenAI
Alleged developer
OpenAI
Alleged harmed parties
Privacy, People whose personal data was included in GPT-2 training data, GPT-2 users, Biometric data subjects

AI governance case studies

For forensic AI governance failure analysis (TAIMScore™ case studies), browse Human Signal’s Failure Files™.

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/357

Data source

Incident data is from the AI Incident Database (AIID).

When citing the database as a whole, please use:

McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. In Proceedings of the Thirty-Third Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-21). Virtual Conference.

Pre-print on arXiv · Database snapshots & citation guide

We use weekly snapshots of the AIID for stable reference. For the official suggested citation of a specific incident, use the “Cite this incident” link on each incident page.