Lawyers for The New York Times and Daily News, who are suing OpenAI for allegedly scraping their works without permission to train AI models, claim that OpenAI engineers mistakenly erased data relevant to the case.
Earlier this fall, OpenAI agreed to provide two virtual machines for counsel to search for copyrighted content within its AI training sets. Virtual machines are software-based computers used for testing, backing up data, and running apps. In a letter filed in the U.S. District Court for the Southern District of New York, attorneys for the publishers state that they’ve spent over 150 hours searching OpenAI’s training data since November 1.
However, on November 14, OpenAI engineers accidentally deleted the publishers’ search data on one of the virtual machines. Although attempts were made to recover the data, the folder structure and file names were irretrievably lost, rendering the recovered data unusable for identifying copied articles in OpenAI’s models.
The counsel for The Times and Daily News had to recreate their work from scratch, resulting in wasted time and effort. The incident highlights the need for OpenAI to search its datasets for infringing content using its own tools.
OpenAI has not provided a statement regarding the incident.
In response to the letter from the publishers’ lawyers, OpenAI’s attorneys denied deleting any evidence and attributed the issue to a system misconfiguration that was requested by the plaintiffs. OpenAI maintains that training models using publicly available data falls under fair use.
While OpenAI has secured licensing deals with various publishers, it has not confirmed whether it trained its AI systems on specific copyrighted works without authorization.
Update: Added OpenAI’s response to the allegations.
