Ziff Davis sues OpenAI, demands destruction of ChatGPT models

If you have ever checked a gadget review on CNET or looked up a game walkthrough on IGN, you have read Ziff Davis. Now, the digital media heavyweight is taking OpenAI to court.

In a lawsuit filed in the U.S. District Court in Delaware, Ziff Davis accused OpenAI of massive copyright infringement, DMCA violations, unjust enrichment, and trademark dilution. The publisher owns over 45 digital brands, employs more than 3,800 people, pumps out nearly 2 million articles a year, and averages over 292 million monthly visits. That makes it one of the largest media networks to sue OpenAI yet.

Scraping ignored warnings and stripped credits

The filings lay out a detailed paper trail. Ziff Davis alleges OpenAI’s GPTBot scraper routinely blew past robots.txt files designed to block content harvesting. Even wilder: after Ziff Davis fired off a cease-and-desist letter in May 2024, crawler activity actually ramped up instead of slowing down.

The complaint also claims OpenAI used specialized software to wipe copyright notices from ingested text. As evidence, Ziff Davis pointed to OpenAI’s publicly released WebText dataset sample, where it identified hundreds of verbatim full copies of its articles. Across training sets, the publisher claims its content shows up millions of times at rates far beyond its footprint on the open web.

In offline tests, ChatGPT reportedly reproduced exact passages from Ziff Davis articles word for word, showing the text was baked directly into the model's weights. Online, the bot pulled Ziff Davis content directly. The publisher also called out hallucinations, pointing to instances where the AI generated fake links, invented quotes, and falsely slapped Ziff Davis brands on made-up facts.

Before filing, Ziff Davis tried to negotiate a licensing deal, but OpenAI turned them down. Now, Ziff Davis wants damages, an injunction, and something far more drastic: the court-ordered destruction of every dataset and AI model containing its work. OpenAI spokesperson Jason Deutrom pushed back, saying models are trained on publicly available data and grounded in fair use.

Why it matters

This lawsuit puts OpenAI’s media strategy under intense pressure.

Right now, OpenAI faces more than 15 legal battles with copyright holders, including The New York Times. While OpenAI managed to sign deals with publishers like Vox Media, The Atlantic, and the Financial Times, journalism researcher Jeff Jarvis described those non-transparent checks as little more than hush money. When OpenAI refuses to pay a massive publisher, it ends up with another existential lawsuit.

For creators and publishers, Ziff Davis is highlighting how easily standard web boundaries like robots.txt are reportedly swept aside when training runs need raw data. If technical guardrails fail and models replace referral traffic with direct answers, the entire ad-and-affiliate revenue model for review sites collapses.

We think the request to erase the models is the biggest ticking bomb here. You cannot easily surgically untrain specific texts from a multi-billion parameter model. If courts take the erasure demand seriously, OpenAI’s fair use defense will not just be about cash settlements; it will be about whether its core engines can remain running as built.