More than $250 million is the damages figure in the USA Today Co. The filing says OpenAI copied 'hundreds of thousands' of articles to train AI models. That number does not prove the training set size, the exact article count, or the method used to identify copied text.

The desk reads the filing as a constraint on OpenAI's data sourcing, not a measurement of model quality. It puts a price on publisher content that the company did not license, and it makes provenance a buyer question. The damages ask is a claim, not a ledger. It does not show which model used the text or when that happened.

The document leaves out the court docket, court name, exact article count, training set size, technical method, and whether the figure includes lost revenue, statutory damages, or future harm. No figure in the filing measures the model's output or the training corpus.

The suit constrains OpenAI's sourcing by attaching a damages claim to unlicensed publisher content. The ask does not prove model quality, training-set size, or the technical path from article to model.

The page leaves out the court docket, court name, exact article count, training set size, technical method, and damages composition.

Treat the $250 million ask as a legal exposure marker, not a training-data audit. Provenance checks for publisher content become necessary, but the corpus, model, and copy-detection method remain unidentified.