The DOJ's OpenAI Brief Is Just One Front in a Much Bigger Copyright War

Sep 7

5 min read

The Battle Across the Arts

On September 1, 2026, the U.S. Department of Justice filed a 20-page brief in The New York Times v. Microsoft Corp. and OpenAI, a copyright infringement case that's been working through a federal court in Manhattan since December 2023. For the first time, the federal government formally took a side — and it sided with the AI companies.

What the DOJ actually argued

The brief doesn't spend much time on the traditional fair-use factors courts usually weigh — purpose, nature, amount used, market effect. Instead, it leans heavily on an industrial-policy argument: the United States (U.S.) has a strategic interest in maintaining "a robust and competitive artificial intelligence industry," and a strict, plaintiff-friendly ruling on copyright would work against that interest.

The more pointed part of the argument is about market structure. The DOJ suggests that if courts require AI companies to license every piece of copyrighted training data, only the largest, best-capitalized firms will be able to afford it. Smaller AI companies and startups — the ones actually driving competition — would be effectively locked out, ceding the field to a handful of incumbents. In other words: strict copyright enforcement, the government argues, doesn't protect creators so much as it protects Big AI's biggest players.

Why The New York Times sees it differently

The New York Times (NYT) and a number of publishers have pushed back hard on this framing. Their position is straightforward: the training data in question is their journalism — reporting that costs real money to produce — and letting AI companies use it without compensation doesn't spur competition, it just transfers value from newsrooms to some of the most valuable companies in the world. Critics of the DOJ's brief have also pointed to timing: it landed amid broader discussions about AI governance ahead of a visit from Chinese President Xi Jinping, adding a geopolitical layer to what's ostensibly a copyright dispute.

Why this matters beyond this one lawsuit

Fair use in the U.S. is decided case by case, and courts pay attention to how other branches of government characterize an issue — even when that characterization isn't binding. This brief effectively puts "AI competitiveness" and "national security" into the fair-use conversation, alongside the traditional factors. That's new, and it's likely to show up in arguments in other AI copyright cases already working through the courts.

This isn't just a news-media problem

It's easy to read the NYT case as a story about journalism specifically, but the same fair-use questions are working through courts on behalf of nearly every kind of creative professional.

Novelists have their own version of this fight, and it's arguably further along. Authors Guild v. OpenAI, filed in December 2023, consolidates claims from sixteen prominent novelists — including George R.R. Martin, John Grisham, Jodi Picoult, Michael Connelly, Jonathan Franzen, and David Baldacci — who allege OpenAI trained its models on their books without permission. The case took a notable turn in October 2025, when the court ruled that short, AI-generated plot summaries of the plaintiffs' novels "may infringe if they capture protected expression at a non-trivial level of detail" — even with zero verbatim quotation. That's a meaningfully different and arguably broader theory of harm than "you trained on my work without paying me": it targets what the model outputs, not just what is ingested, and it's among the first U.S. rulings to do so.

Anthropic, facing similar claims from authors over pirated books used in training, chose a different path than litigating fair use to a verdict: in 2025 it agreed to pay $1.5 billion to settle the case, the largest publicly reported copyright recovery in U.S. history. That settlement didn't resolve the underlying legal question — it sidestepped it — but it set a real-dollar case, and every AI company's legal team, is now aware of.

Screenwriters and visual artists are represented in this fight too, just through their studios. Disney, NBCUniversal, and DreamWorks sued the AI image generator Midjourney, arguing it trained on and now reproduces the copyrighted characters, scenes, and stylistic work behind their franchises — calling the platform "a bottomless pit of plagiarism." And in music, Concord Music Group has brought a comparable claim against Anthropic over song lyrics, on behalf of songwriters and music publishers.

Put together, that's a striking pattern: journalists, novelists, screenwriters, and songwriters are all litigating close variants of the same underlying question — does training an AI model on copyrighted creative work, and generating outputs derived from it, require a license — and getting different outcomes depending on which court, which defendant, and which legal theory applies. A poet or independent author with a handful of books in the same training set as these plaintiffs has no realistic path to litigate individually, but stands to be affected by whichever precedent eventually holds.

What this means depending on where you sit

If you build AI products, the DOJ's brief is a tailwind — but it's not a resolution. The case isn't decided, and a favorable government brief doesn't guarantee a favorable ruling, especially with a competing legal theory (the Authors Guild's output-level infringement argument) gaining traction in a different case. Data provenance and licensing practices are still worth documenting carefully, regardless of how the NYT case lands.

If you create or publish any kind of copyrighted works — articles, books, scripts, lyrics, or illustrations — the uncertainty here is exactly why licensing deals, rather than litigation, have become the more common path for larger rights-holders dealing with AI companies. Anthropic's settlement shows what that looks like at scale; watching how the NYT and Authors Guild cases resolve will shape whether direct litigation becomes more or less attractive for the writers and creators who haven't yet struck a deal.

If you're in-house counsel, legal ops, or represent creative professionals in any capacity, this is a good moment to revisit how data provenance is documented for any AI tools your organization builds or licenses — and to note that liability theories are now extending beyond training data to the outputs a model generates, which is a meaningfully different compliance question.

We'll be watching how the court responds to the DOJ's brief, how the Authors Guild's output-infringement theory develops, and what both signal for the dozen or so similar AI copyright cases still working through U.S. courts on behalf of writers, artists, and musicians alike.

Not sure how this affects your content, your product's training data, or your legal exposure? We help teams sort through exactly this kind of ambiguity — data provenance reviews, licensing strategy, and plain-English guidance on where the law actually stands today. Book a free consultation call with us and we'll help you figure out where you stand.

📅 Book online: www.atmoip.com/book-online

🌐 Visit: www.atmoip.com

📩 Email: info@atmoip.com

Have a take on where this is headed? Drop it in the comments — we read every one.

You can also follow aTMospheric IP (@atmoip) for practical insights and updates.

Facebook

Instagram

Bluesky

Protect what makes your business unique.