Breaking

Unsealed Filings Show OpenAI Staff Weighed Book Costs as Copyright Fight Widens

September 19, 2026 at 07:16 EDT

Foundation Models
Policy & Regulation

Newly unsealed court material says OpenAI employees discussed whether to buy books for early ChatGPT training, while also weighing the legal and reputational risks of using pirated book collections. The disclosures add new pressure to OpenAI and Microsoft as authors and publishers argue that major AI systems were built on copyrighted work without permission.

Sep. 2026 · OpenAI

Book data moves to the center of AI copyright

Unsealed filings say OpenAI staff weighed the cost of buying books while discussing risky sources for early ChatGPT training. The fight now turns on fair use, market harm and what companies knew when models were built.

$1.5B

Anthropic settlement benchmark in a separate books case

465,000

Books cited in the Anthropic dispute

2023

Year authors filed major claims against OpenAI

Scale of the books dispute

465,000 books

10 blocks, about 46,500 books each

The OpenAI filings add pressure because a separate books case has already set a high-dollar reference point for alleged use of pirated books.

AI industry view

Large-scale text training is described as essential to capable language models, with companies arguing that training is transformative.

Creator concern

Authors and publishers argue unauthorized books and articles can create direct market harm when chatbots summarize, imitate or replace source works.

The legal chain

Data source→Training use→Market effect→Fair-use ruling

The filings shift attention from broad doctrine to internal knowledge, available alternatives and whether early training choices affect liability.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.