Senior Partner Eleanor Vance scanned the latest brief in her office at Sterling & Hayes. Her firm was representing the Global News Collective, a group of big news publishers with a big problem: their content was being hoovered up by AI models. It’s 2026, and the fight over AI and copyright is getting nasty. Publishers want to be paid and they want clear rules, and the effective legal strategies are only just starting to take shape.
Key Takeaways
- News publishers are suing AI developers for copyright infringement over the mass, unauthorized scraping of their articles for training data.
- Licensing deals are becoming the go-to proactive move, with publishers offering tiered access and putting tight restrictions on usage.
- The whole fight in court often comes down to one thing: is training an AI a “far-reaching use” or just straight-up copying?
- Publishers are getting serious about proving where AI training data comes from, using tech to link scraped content directly back to their work.
- News organizations are pushing hard in D.C. for new laws, specifically to update things like the Digital Millennium Copyright Act for the AI era.
The Unauthorized Ingestion: A Publisher’s Nightmare
Eleanor’s desk was a mess of forensic data reports. They showed exactly how huge chunks of the Global News Collective’s archives, some articles decades old, had been scraped by major AI models without a single permission slip. The value of original journalism itself is being eroded, she’d told her clients. The first lawsuits, filed back in late 2024 in the Southern District of New York, were built on a simple claim of copyright infringement under 17 U.S. Code § 106. The argument was that AI developers copied protected works to train their models and in doing so created unauthorized derivative works (the AI’s output). In Eleanor’s mind, this was a clear violation of existing law.
A particularly bad case involved an AI spitting out full summaries of the Collective’s investigative series on Atlanta’s municipal corruption. Even though the AI rephrased things, it used the exact same factual details, the specific angles, and even quotes that the Collective’s own journalists had dug up. “Look at this,” Eleanor told her junior associate, David Chen, showing him a side-by-side. “The AI isn’t learning facts. It’s learning our voice, our research methods, our storytelling. That’s our property.”
The Fair Use Battleground: Far-reaching Use vs. Derivative Works
The AI companies’ defense always comes back to fair use. Their main argument is that feeding copyrighted material into a model for training is a far-reaching use. They say the models aren’t just copying articles, they’re learning patterns from the data to create something totally new. This whole argument relies on the Supreme Court’s take on far-reaching use from cases like Campbell v. Acuff-Rose Music, Inc., where the court asks if the new work adds “new expression, meaning, or message.”
Eleanor’s team shot back that just rephrasing or summarizing an article isn’t far-reaching, especially when the AI’s output is directly competing with the original story. “If an AI can write a news story that a reader thinks is from a human journalist, and that story is based completely on our reporting, how is that far-reaching?” Eleanor asked in a deposition. “It’s just sophisticated content laundering.” Her team is hammering on the fourth fair use factor: the effect on the market for the original work. When an AI-generated story replaces a human-written one, the market for real journalism takes a hit, a point the News Media Alliance has made in its own papers on AI.
Proactive Measures: Licensing and Data Provenance
Publishers aren’t just waiting to sue. They’re getting proactive. Many, including some in the Global News Collective, are now offering strict licensing agreements to any AI developer who wants to use their archives. We’re seeing contracts with tiered pricing based on data volume, the model’s purpose, and firm limits on monetizing the AI’s output. A Reuters analyst recently pointed out that these deals can specify data caps, demand attribution when the AI leans heavily on a source, and even require revenue sharing for some commercial uses.
On top of that, publishers are pouring money into tech that tracks the provenance of their data. Digital watermarks, better metadata tagging, and even blockchain registries are becoming standard practice. This tech gives publishers the receipts to prove their content was ingested without permission. As Eleanor noted, the burden of proof is shifting. AI companies can’t just claim they didn’t know where the data came from anymore. Publishers are building undeniable digital trails, which helps them prove direct infringement and shut down any claims that the AI came up with the ideas on its own.
Legislative Advocacy: A Call for Clearer Lines
Court battles are slow and they cost a fortune. So news publishers are also pushing for new laws. In D.C., groups like the National Music Publishers’ Association and the News Media Alliance are lobbying to amend laws like the Digital Millennium Copyright Act (DMCA) to finally deal with AI training data. They’re asking for things like mandatory disclosure of training sets from AI developers, required licensing frameworks, and new penalties for AI-driven infringement. As Eleanor put it during an ABA panel last spring, the existing laws were written before anyone imagined AI training and fail to protect creators from having their entire archives scraped. She argued that Congress needs to draw clear lines so the courts aren’t stuck trying to apply old laws to this new problem.
A key part of this legislative push is to clarify what “fair use” even means for AI training. Publishers want a new rule: if you’re using copyrighted stuff for a commercial AI without permission, and that AI competes with the original, the use is presumed *not* to be fair. This would flip the script completely. Instead of publishers having to prove infringement, the AI companies would have to prove their use was fair. It’s a huge change, and you can bet the tech industry is fighting it tooth and nail.
The Global News Collective’s Next Move
Back in her office, Eleanor put the finishing touches on a new motion to compel discovery. The Global News Collective wanted to see the exact datasets a defendant AI company used, aiming to pinpoint every single article they ingested, down to the file and timestamp. This kind of forensic evidence would show the true scale of the infringement and blow up any arguments about minimal impact. “We’re not just trying to win one case,” Eleanor told David. “We’re trying to set a precedent. If AI companies can just take whatever they want, the whole model for funding credible local news funding collapses.” Her firm was also starting to look into class-action suits, since most smaller publishers can’t afford to fight these battles alone.
What are the primary legal claims news publishers are making against AI companies?
They’re suing for copyright infringement. The core claim is that AI developers copied and used their articles to train AI models without permission, which counts as reproducing protected work and creating unauthorized derivative works.
How do AI companies defend against copyright infringement claims?
Their main defense is fair use. They argue that training an AI is a “far-reaching use” because the model isn’t just storing copies but is learning patterns to create something new.
What proactive strategies are news publishers employing to protect their content from AI?
Instead of just waiting to sue, publishers are creating licensing deals with tiered pricing and clear rules for AI developers. They’re also using tech like digital watermarks and metadata to track exactly where their content ends up.
Are there legislative efforts underway to address AI and copyright?
Yes. News organizations are pushing Congress to amend laws like the DMCA. They want new rules that would force AI companies to reveal their training data and create mandatory licensing systems.
What is the central debate regarding “far-reaching use” in AI copyright cases?
It all comes down to this: are AI models just creating sophisticated summaries of existing work (a derivative work) or are they producing something genuinely new that has a different expression or meaning (a far-reaching use)?