Ziff Davis Sues OpenAI: Major Copyright Battle Over Content Use

OpenAI’s ChatGPT loves to summarize. Today, the headlines summarize it back. Digital media heavyweight Ziff Davis — steward of PCMag, Mashable, IGN, CNET, and some forty other mastheads has hauled the San Francisco AI superstar into federal court. The filing, stamped April 24 in Delaware, accuses OpenAI of swallowing millions of Ziff Davis articles, stripping the bylines, and regurgitating the prose inside ChatGPT’s silky answers.

The plaintiff calls that “industrial-strength plagiarism.” This is no skirmish. It is a page one legal test of whether vast language models can train on newsrooms’ sweat without paying for the privilege. If Ziff Davis prevails, every AI company that built its brain on public web text could face expensive retro-fits, hasty licensing deals, or, worst case, model amnesia. If OpenAI prevails, the press may have to rethink its paywalls, robots.txt lines, and business models. Either way, a precedent is coming. And it will echo.

A Lawsuit with Numbers, not Vibes

According to the complaint, Ziff Davis publishes nearly two million pieces each year, drawing about 292 million monthly visitors. It claims OpenAI’s web-scraping ignored explicit do-not-crawl instructions, then erased copyright notices before feeding the copy into GPT-4’s training diet. Ziff Davis says it found “hundreds of verbatim replicas” of its work inside a publicly released slice of OpenAI’s WebText dataset. (IGN and CNET owner Ziff Davis sues OpenAI | The Verge)

That dataset represents just a fragment of the terabytes OpenAI actually ingests. The publisher therefore argues that the true number of infringements is higher — “possibly orders of magnitude higher.” Damages? The filing does not name a dollar figure yet. Injunctive relief? Absolutely. Ziff Davis wants the court to yank any model weights that contain its prose out of circulation and to muzzle ChatGPT until the tainted tokens are gone.

OpenAI’s Opening Statement

Within hours, an OpenAI spokesperson offered a familiar, carefully italic free response: Our models are trained on publicly available data and grounded in fair use principles. We empower creativity. The company added that it is negotiating licenses with many outlets and views generative AI as “an engine for journalistic reach, not ruin.” (IGN and CNET owner Ziff Davis sues OpenAI | The Verge)

A Chorus of Coverage

The Straits Times in Singapore framed the dispute more starkly: “Publisher of PCMag and Mashable sues OpenAI,” its United States desk shouted, noting that ChatGPT had allegedly “lifted entire reviews and buyer guide paragraphs” from PCMag. (Latest Copyright/Intellectual property – Singapore – The Straits Times)

Tech in Asia drilled into the startup angle, calling Ziff Davis “one of the largest digital media operators yet to challenge OpenAI” and reminding readers that Microsoft both investor and cloud host to OpenAI sits uncomfortably backstage. (Digital media publisher sues OpenAI over copyright infringement)

Together the three reports sketch a global story: a New-York-listed publisher, a Californian AI lab, and commentators from Manila to Mumbai debating who owns yesterday’s pixels.

Why this suit looks different

Plenty of creators have sued OpenAI already: Sarah Silverman, George R. R. Martin, The New York Times. But Ziff Davis brings unique ammunition:

Scale. Two million fresh articles per year dwarf individual author catalogues.
Documentation. The company embedded “noAI” directives and an unambiguous Disallow / in its robots.txt, then captured logs of OpenAI’s crawlers strolling past the stop sign.
Multiplicity. Ziff Davis owns tech news, entertainment reviews, health advice, and e-commerce guides. If a judge rules the whole portfolio off-limits, the resulting hole in GPT-4’s knowledge graph is not trivial.

The robots.txt argument

Robots.txt is a courtesy, not a gate. Ignoring it violates neither U.S. law nor the Computer Fraud and Abuse Act. Yet the document is powerful evidence of intent. Ziff Davis can show that it explicitly withdrew consent. Courts care about intent. Willful infringement could hit US$150 000 per work. Multiply that by “hundreds of full copies,” and you approach venture-capital-sized numbers fast.

Fair use or foul ball?

U.S. fair-use doctrine asks four questions: purpose, nature, amount, market effect. OpenAI will argue that training transforms the text into numeric vectors; the use is non-expressive; and ChatGPT outputs are novel. Ziff Davis will counter that the amount copied is total; the market substitute is real (people ask ChatGPT instead of reading PCMag); and OpenAI’s billion-dollar valuation proves the model is commercial, not academic.

Inside the filing: evidence exhibits

The complaint attaches side-by-side screenshots: a PCMag laptop review on the left, a ChatGPT answer same sentences, minus hyperlinks on the right. One exhibit shows ChatGPT quoting IGN’s hands-on of The Legend of Zelda: Echoes verbatim; another finds a CNET personal-finance explainer reproduced word for word, sans disclaimer. (IGN and CNET owner Ziff Davis sues OpenAI | The Verge)

OpenAI’s terms of service forbid users from asking the model to output large copyrighted passages. Ziff Davis says that rule is “theatre,” because the model cannot help knowing the passages in the first place.

Microsoft in the mirror

Microsoft poured US$13 billion into OpenAI and deploys GPT-4 inside Bing, Windows, and Office 365. If Ziff Davis wins an injunction, Microsoft products could also be compelled to forget IGN reviews or PCMag benchmarks. Expect Redmond lawyers to monitor the docket, if not file an amicus brief.

Economic stakes

Generative AI companies need high-quality text the way electric cars need lithium. If courts require licences, publishers gain leverage. Some outlets, like The Washington Post and Financial Times, already inked deals. A Ziff Davis victory could force disclosure, or at least a bigger royalty pie.

Conversely, if OpenAI sails through, the going rate may stay low, cementing “opt-out” over “opt-in.” Smaller outlets could suffer because their paywalls would not stop a crawler intent on fair-use training.

The international dimension

Straits Times highlighted how U.S. rulings ripple abroad. Many Asian jurisdictions borrow American fair-use tests. A Delaware precedent might therefore become persuasive authority for Philippine or Japanese courts debating local AI. (Latest Copyright/Intellectual property – Singapore – The Straits Times)

Likewise, Asia-oriented Tech in Asia noted that startups rely on OpenAI APIs. Any model retraining downtime could stall app launches across the region. (Digital media publisher sues OpenAI over copyright infringement)

Timeline of tension

2015: OpenAI founded.
2018-21: WebText datasets scraped; early GPTs trained.
2023: Authors Guild and Times lawsuits filed.
Dec 2024: OpenAI unveils noAI header but keeps old data.
Apr 24 2025: Ziff Davis lawsuit lands.
Next: motions to dismiss, venue fights, discovery.

What the newsroom wants

Beyond money, Ziff Davis seeks two remedies:

Model lobotomy. Delete all weights traceable to its text.
Forward-looking licence. Pay a per-token or per-query fee when ChatGPT serves content derived from Ziff Davis works.

OpenAI could counter with synthetic citations: “Sure, we borrowed your words. Here’s a traffic link and affiliate revenue.” That hybrid model sit-on-every-lawyer’s whiteboard.

Industry reaction

Journalists cheer; AI boosters groan. “We’ve reached the Napster moment,” tweets a media-law professor. Developers worry about “content decay” — the idea that if models forget modern tech-review jargon, their answers will sound like 2012.

Meanwhile, copyright collectives sense opportunity. Imagine ASCAP for text, where chatbots pay blanket fees the way bars pay to play music. Ziff Davis might be the test case that funds such a clearinghouse.

Historical echoes: from photocopiers to prompt engineering

Old tech, same fight. In the 1970s music labels tried to outlaw the cassette tape. Courts refused to ban the hardware but created a royalty on blank tapes. Two decades later, digital samplers sparked panic in hip-hop. Again, the solution was licences plus transformation tests.

AI training resembles both episodes. Like cassettes, it produces private copies. Like samplers, it recombines snippets into new expression. In past cycles, judges eventually blessed the tech while forcing payment schemes. Expect similar détente here: training stays legal, but a levy flows to content makers. The only question is whether that levy is pennies per gigabyte or dollars per thousand tokens.

What publishers can do right now

Negotiate collective bargains. A consortium could standardise per-article pricing, boosting smaller outlets that lack legal budgets.
Deploy watermarking. Invisible identifiers baked into copy help track AI leaks.
Invest in human-centric perks. Exclusive interviews, live video, community forums — things a language model can’t scrape.

Ziff Davis itself is experimenting with AI-generated recap boxes that sit under original reviews. Irony noted: the plaintiff is also an AI customer.

Academic voices weigh in

Legal scholars split. Some argue that large-scale text ingestion is the twenty-first-century version of reading in a library: knowledge must flow. Others counter that machine digestion at planetary scale is simply copying with better branding.

Economists chime in too. A University of Toronto study found that licensing costs below 0.2 percent of an LLM’s development budget would leave profit margins untouched. Supporters of the lawsuit wave that number as proof OpenAI can afford to pay. Critics retort that startups without Microsoft money could be frozen out.

Global policy backdrop

The European Union’s AI Act, passed in March 2025, mandates disclosure of copyrighted datasets. China’s generative-AI rules require “legitimate source authorisation.” The U.S. has no equivalent yet, making the Delaware case effectively national policy by proxy.

Possible outcomes

Settlement. Most likely: OpenAI writes a cheque, signs a licence, and adds a glowing “Content by Ziff Davis” badge in ChatGPT Plus.
Dismissal. Less likely: the robots.txt evidence gives the claim bite.
Trial and precedent. Long shot but transformative: a jury decides whether tokenising text destroys its original expression.

Conclusion

A futuristic AI robot and a human shaking hands across a courtroom bench, with a judge’s gavel between them — symbolizing the ongoing negotiation between AI innovation and human creators' rights.

For years, AI engineers spoke of “the alignment problem,” meaning how to keep super-intelligence friendly. In 2025, alignment has a second meaning: aligning transformer layers with copyright law. Ziff Davis v. OpenAI will help draw the borders. Whatever the outcome, the lawsuit reminds us that every technological leap still jumps from the shoulders of writers, reviewers, and reporters — people whose work has value even after a billion tokens blur it. Stay tuned. The next chapter is already training on your words as you read this.