US Government Backs OpenAI in NYT Copyright Fight

On 2 September 2026 the Trump administration filed a 20-page brief in The New York Times Company’s lawsuit against OpenAI, defending the ChatGPT maker’s unlicensed use of copyrighted works to train large language models. TechCrunch reported that the filing casts that training as fair use and ties the outcome to whether the United States keeps global leadership in artificial intelligence — language that echoes a Trump executive order on AI competitiveness.

What the administration brief actually argues

TechCrunch’s report on the filing quotes the brief saying the United States has a strong interest in a robust, competitive AI industry that sets global practice, and that constraining LLM development under a misunderstanding of fair use would hinder creative and scientific progress plus American prosperity. Publishers including the Times argue that scraping copyrighted articles and books without permission is infringement; OpenAI and peers argue training is transformative because models learn patterns rather than republish works wholesale.

The brief does not decide the case. Jurisdiction sits with the Southern District of New York, and the administration’s authors lack power to enter judgment. Still, a White House-aligned intervention on fair use can shape amicus weight, settlement leverage, and how other agencies talk about training data. For readers following Europe’s harder product rules, PromptCrates’ EU AI Act enforcement powers coverage shows how capability and transparency duties already bite on a different clock than US copyright doctrine.

How prior training cases have landed

TechCrunch notes that AI-training copyright fights have largely favored model builders so far. A key data point is Anthropic’s roughly $1.5 billion settlement with writers ordered by Judge William Alsup: the pain centered on using illegal shadow libraries to obtain books, not on the act of training as such. Alsup compared LLM training to a reader aspiring to write differently rather than to race ahead and replace the original — language OpenAI’s supporters will cite, and publishers will try to distinguish.

That distinction matters for product teams. If courts keep separating piracy of inputs from transformative training, licensing markets may focus on clean corpora and provenance rather than blanket bans. If judges reject fair use for commercial LLM training, subscription newsrooms and book houses gain bargaining power overnight. Either path reshapes how startups scrape the open web and how enterprises audit vendor data sheets.

Why this brief still carries weight

Even without a ruling, the filing signals that US industrial policy and copyright litigation are colliding in the same courtroom. OpenAI’s broader legal and security pressure — including trade-secret fights covered in PromptCrates’ Apple–OpenAI Chang Liu evidence story — already shows frontier labs defending capability on multiple fronts. A government brief that frames fair use as national competitiveness raises the political temperature for any SDNY outcome.

Publishers will argue that unlicensed training hollows out the market for journalism and books; AI firms will argue that requiring licenses at training scale is impractical and would cede leadership abroad. Courts still have to apply the four fair-use factors to specific facts, including whether outputs substitute for Times articles. Until then, treat the brief as a loud policy marker, not a green light to ignore licensing diligence.

Enterprises buying generative tools should ask vendors how training corpora were sourced, whether shadow libraries appear in the chain of custody, and how output filters reduce verbatim regurgitation of paywalled text. Those questions remain relevant whether or not SDNY ultimately sides with OpenAI. The administration’s 20 pages change the narrative weather; they do not rewrite copyright statutes.

Internationally, the brief may widen the gap between US fair-use optimism and stricter European or Asian regimes that emphasize consent and remuneration. Multinationals shipping one model worldwide already juggle conflicting expectations; a US government thumb on the scale toward permissive training strengthens the case for region-specific data policies. Watch how other agencies and state attorneys general react in coming weeks.

For newsrooms and authors, the practical risk is asymmetric: if fair use holds, collective licensing may still emerge as a commercial peace treaty; if it fails, injunctions and statutory damages become the lever. For OpenAI, Claude, and Gemini builders, the brief is helpful optics heading into discovery fights over what was ingested. The SDNY docket — not a press quote — will decide who pays whom.

Litigation calendars move slower than model release cycles, which is why briefs like this matter between hearings. Investors read them as policy weather; newsroom unions read them as leverage forecasts; open-source maintainers read them as a hint about whether US training norms will diverge further from Europe. A single SDNY opinion can still surprise everyone, but the administration’s framing locks fair use to industrial strategy in a way that will appear in future filings.

OpenAI and the Times still have discovery, experts, and remedy fights ahead. Nothing in the 20 pages freezes scraping practices or forces a compulsory license. What changed on 2 September 2026 is the public coalition: the executive branch is now on record arguing that misunderstanding fair use could blunt American AI leadership. That sentence will be quoted — approvingly or critically — for the rest of the case.

Secondary coverage will also compare this intervention with how other governments brief courts on platform cases. Soft power through amicus-style filings is cheaper than statute, and it can be walked back after an election. For now, publishers should assume US policy elites want training to remain broadly lawful when inputs were obtained cleanly, while still leaving room to punish piracy pipelines like those that sank Anthropic into a billion-dollar settlement.

Sources

• US government sides with OpenAI on issue of training LLMs on copyrighted material — TechCrunch, 2 September 2026