The battlefield moved

Sarah Silverman sued OpenAI after alleging that her memoir, The Bedwetter, had been copied without authorization into training datasets. Three years later, that question almost looks quaint. Her case now sits inside a consolidated New York fight involving authors, publishers and The New York Times. What began as a claim about one writer’s book has become a forensic examination of how the modern AI corpus was built.

Two federal rulings supplied the industry’s emerging doctrine. In Bartz v. Anthropic, Judge William Alsup treated model training as transformative fair use on the record before him—but separated that use from the creation of a permanent library assembled with pirated books. In Kadrey v. Meta, Judge Vince Chhabria ruled for Meta on a thin record while emphasizing the plaintiffs’ failure to show meaningful market dilution.

Dispatch analysis: Training, acquisition and output are becoming distinct legal products. A developer may have a strong defense for what a model learned and still face enormous exposure for how the source material entered the pipeline.

The $1.5 billion warning

Anthropic’s settlement turns provenance from an abstract compliance concern into a line item. The lesson is not that every training use requires a royalty. It is that a fair-use destination does not necessarily cleanse an infringing route. The acquisition record—purchase, license, scrape or pirate archive—may determine the bill.

Where the money is moving

The market is already rewarding clean access. Shutterstock’s Data, Distribution and Services revenue reached $203.3 million in 2025, up from $137.3 million in 2023. Reddit reported $140 million of “other revenue” in 2025, a category that includes content licensing but is not broken out separately. These are not standardized royalties. They are early evidence of a wholesale market for legally usable human knowledge.

Dispatch analysis: The premium asset is becoming copyright plus access plus provenance plus machine-readable rights. Large catalogs and data intermediaries gain leverage; individual creators risk retaining rights without negotiating scale.