Microsoft executive calls company's AI scraping practices 'theft,' filing reveals
What's the story
Newly unsealed documents in a copyright lawsuit filed by The New York Times against OpenAI and Microsoft have revealed shocking admissions about the companies' artificial intelligence (AI) scraping practices. A top Microsoft executive privately described these practices as "theft," while OpenAI's leadership admitted that their AI models posed an "existential threat" to publishers and journalists whose work was used for training.
Scraping tactics
Allegations of mass scraping and paywall bypassing
The unsealed documents detail how OpenAI and Microsoft allegedly scraped content from The New York Times without permission.
They are accused of bypassing paywalls undetected, building training datasets through mass scraping, and deliberately stripping copyright notices from the data.
This revelation contradicts OpenAI's fair use defense in the lawsuit, especially the rule's requirement that use doesn't harm or substitute for the original work's market.
Market impact
Microsoft's internal presentation on Copilot's impact
Microsoft's own data shows its Copilot "answer engine" reduced click-through rates for The New York Times's domain by up to 93% compared to traditional Bing search.
An internal Microsoft presentation from January 2024 described this decline as a "doom loop" that would "hurt the performance of our models and the entire web at the same time."
Licensing stance
Nadella's testimony on paywalled content usage
Microsoft CEO Satya Nadella testified earlier this year that "anything that is paywalled should be licensed by anyone who wants to use it...for grounding or training."
He also said if he had known OpenAI had scraped and trained on information behind a paywall, he would have "invoked [Microsoft's right to] require OpenAI to retrain its models."
AI risk
OpenAI acknowledged potential threat to publishers
OpenAI's Head of ChatGPT, Nick Turley, admitted in an internal communication that products like the chatbot pose an "existential threat" to publishers.
He described them as "largely substitutive" and said they will become more so as they improve.
This admission highlights how these technologies could directly compete with original work rather than just transform it.
Employment risk
Generative AI could disrupt employment, warns Microsoft document
A Microsoft document warned of a "real risk" that generative AI could "significantly disrupt the employment of the very people who generated the data on which the foundation model was trained."
This revelation highlights potential job losses in industries where these technologies are deployed.
The scale of copying is also massive, with OpenAI's mid-training datasets containing over 91,692 copies of works published by NYT and others.
Data sharing
OpenAI and Microsoft collaborated on training data for each other
The unsealed documents also revealed that OpenAI provided the entire GPT-3 training dataset to Microsoft, which it used to evaluate how to implement OpenAI's models in its own commercial products.
Similarly, Microsoft provided training data to OpenAI through initiatives called Project Taxi and Project Mango.
The companies allegedly assembled the Project Mango data into a training dataset containing copies of at least 160,903 unique works from news publishers.