A settlement demand is a probability of winning multiplied by a damages figure, less the cost of getting there, and on September 1 the Justice Department leaned on the first term for every plaintiff in the docket at once, since its statement of interest covers all matters consolidated in MDL 25-md-3143 and not the New York Times case alone [2][3].
The instrument matters more than the rhetoric. Filed under 28 U.S.C. section 517, the government states a position without becoming a party, the filing carries no binding authority, and Judge Sidney Stein may give it whatever weight he sees fit, including very little [4]. Which means the three lawyers who signed it [5][3] face none of the litigation risk that comes with the argument: no discovery to sit for, no exposure if Stein discounts the brief entirely, and the position still works in every conference room where a content licence is being priced. That is a cheap trade in credibility terms, or rather, the more interesting version: the administration has spent executive-branch credibility, which the publisher characterises as industrial policy preference rather than neutral technology-law analysis, in a venue where the downside is a footnote [7][12].
The authority underneath is thinner than the language. Authors Guild v. Google, the 2015 Second Circuit decision upholding the scanning of millions of books for a search index without reproducing them, was eleven years old when the brief invoked it [10][2], and the newer support is a single district judge, Alsup in Bartz v. Anthropic, whose phrase was "quintessentially transformative" applied to lawfully acquired books [11]. Publishers should read the qualifier rather than the adjective, because a fair-use holding on the training step is silent on how the corpus was assembled, and TechTimes points to its own earlier reporting that a deposition showed OpenAI tracking copyright violations it said it could not track [14].
The government's own move was made in a courtroom rather than in a bill. The brief tells writers and journalists to seek compensation from Congress or the licensing table [13], leaving that recommended route without a sponsor. The Times, meanwhile, has been in this for roughly 33 months since December 2023 [1], funding counsel against a premise, the rigorous four-factor test, that now has the executive branch arguing the other side [12].
This runs a few different ways. Stein could treat a non-party filing as what it is and rule on the record, restoring the option value of the MDL. He could cite it, in which case publishers read the signal and licence terms clear at whatever an AI lab thinks convenience is worth rather than at avoided damages. Or the case narrows to acquisition, and the money turns on provenance of files rather than on what the model did with them.
This is probably wrong, but my read is the third: training-as-fair-use is where the government put its weight [8][9], and that is also the part of the case publishers were least likely to win. I would abandon the view if Stein's opinion adopts the brief's framing across all four factors, or if a publisher survives summary judgment on market dilution, either of which puts damages back on the table and the leverage back with the archive.
