The New York Times and The Daily News are accusing OpenAI of withholding critical evidence in a copyright lawsuit over its ChatGPT model. According to the outlets, OpenAI allegedly hid internal data showing it had already searched its training corpus for copyrighted works and maintained a database of 78 million de-identified ChatGPT conversations. These revelations come as part of a broader legal battle over whether OpenAI violated copyright laws by training its AI models on the Times’ journalism. The plaintiffs claim OpenAI made it unnecessarily difficult to access information that the company had already collected, including by redacting key details in a submitted sample of 20 million chat logs.
In an April court-ordered deposition, OpenAI data privacy engineer Vinnie Monaco allegedly revealed that the company had conducted internal searches and evaluations of its training data to identify copyrighted journalism. Monaco’s testimony also indicated that OpenAI had implemented a 'Bloom' filter as part of 'Project Giraffe,' a set of tools designed to detect and record instances of regurgitation in AI outputs. These tools were reportedly introduced shortly after the lawsuit was filed, raising questions about the company’s transparency and cooperation with the legal process.
The plaintiffs argue that OpenAI violated a court order by deleting billions of ChatGPT outputs and substituting millions of logs in the requested sample. They are now seeking judicial intervention to prevent OpenAI from using the 20 million chat log sample as evidence, claiming it is unreliable. OpenAI spokesperson Drew Pusateri denied the allegations, accusing the Times of attempting to access private user data as its case weakens.
Source: techcrunch
