Nonfiction Writers Sue OpenAI, Microsoft for Using Their Books to Train ChatGPT

ChatGPT and nearly every other similar Large Language Model (LLM) AI rely on massive troves of data to work. OpenAI scraped much of its data from the internet, but some of it also comes from sources like nonfiction books, and authors are now seeking compensation in a new lawsuit.

As Reuters reports, author Julian Sancton claims that OpenAI copied tens of thousands of nonfiction books without permission to train ChatGPT. He’s leading a proposed class-action lawsuit, filed in Manhattan Federal Court on Tuesday, that says “Defendants’ models were calibrated (or ‘trained’) by reproducing a massive corpus of copyrighted material including, tens or hundreds of thousands of nonfiction books.”

The lawsuit comes as OpenAI faces several other legal and in-house issues. The company is already the subject of other lawsuits concerning private data, using fictional writers’ content, and even actors like Sarah Silverman. If that’s not enough trouble, OpenAI’s Board fired its CEO last weekend, only to walk back that decision and promise to hire a new board.

This lawsuit differs from others in that it’s the first to include Microsoft along with OpenAI. Microsoft is a heavy investor in OpenAI, and its Bing Chat, AI-powered Search, and new Copilot feature rely heavily on OpenAI technology. It’s almost a Microsoft-flavored ChatGPT. The lawsuit claims Microsoft is “deeply involved” in training ChatGPT, so it is liable for any infringement as well.

While OpenAI and Microsoft aren’t commenting on this latest lawsuit yet, OpenAI has in the past stated that ChatGPT-generated content doesn’t constitute “derivative works” and, as such, doesn’t infringe on copyright.

LLMs have exploded in the past year, and many questions about copyright haven’t been resolved in the courts yet. The lawsuits may go nowhere or force a change in how LLMs compensate creators for data used in training. It’s too soon to tell.