Ziff Davis v. OpenAI

A publisher without a licensing deal

Ziff Davis filed its complaint in April 2025 as, at the time, the most prominent US digital publisher without an existing licensing arrangement with OpenAI. Its brands span consumer tech coverage (PCMag, CNET, ZDNET), entertainment (Mashable, IGN), lifestyle (Lifehacker) and health (Everyday Health) — a broad enough footprint to make the case a useful test of how courts treat robots.txt as an access control.

The complaint alleges OpenAI's GPTBot kept crawling Ziff Davis sites and training on their content even after the company deployed robots.txt directives specifically blocking that bot — and that bot activity on its sites spiked rather than stopped. The case was moved into the OpenAI multidistrict litigation in New York within a month of filing.

The robots.txt ruling

On December 15, 2025, Judge Stein issued the ruling that gives this case its most cited line. He dismissed Ziff Davis’s claim that ignoring robots.txt violated the DMCA’s anti-circumvention provision, holding that a crawl directive is not a technological protection measure — comparing it to a sign on a lawn rather than a lock on a door.

That holding cuts a specific, narrow path. It kills the anti-circumvention theory for robots.txt violations generally, not just for Ziff Davis. But it did not kill the case. Stein simultaneously advanced the contributory-infringement and DMCA copyright-management-information removal claims — the theories built on OpenAI actually reproducing or stripping attribution from copyrighted content, not merely on crawling past a disallow line.

“A robots.txt file no more controls access than a "keep off the grass" sign controls a lawn.”— Order on motions to dismiss, S.D.N.Y. (MDL No. 3143), Dec. 15, 2025 (Judge Stein)

What survived, what didn’t

Alongside the Section 1201 dismissal, Stein also dismissed Ziff Davis’s unjust-enrichment claim and narrowed trademark dilution down to the Mashable mark specifically, rejecting broader dilution theories across the company’s other brands. He also stayed discovery into newer model generations — o1, o3, GPT-4.1, GPT-4.5, GPT-5 and others — not otherwise already at issue in the consolidated MDL.

What remains live: direct and contributory copyright infringement, and the DMCA Section 1202 claim over removal of copyright-management information. Those are the theories that will define whatever comes next in the case.

A new sanctions fight

On July 9, 2026, Ziff Davis joined a sanctions motion filed by newspaper plaintiffs in the broader MDL, alleging OpenAI concealed and destroyed discovery evidence. That motion is pending and separate from the merits rulings — but it signals the litigation has moved from pleading fights into a contentious discovery phase, with accusations aimed at OpenAI’s conduct in the litigation itself, not just its training practices.

The lesson for publishers on technical controls

The headline lesson for anyone relying on robots.txt as protection: it isn’t one, legally. A disallow line signals intent, and it may support other claims — as it apparently does here for the CMI-removal theory — but it carries no anti-circumvention teeth on its own. Publishers who want enforceable technical protection need actual access controls: authentication, rate-limiting, or licensing terms with contractual teeth, not just a text file bots can choose to ignore.

The model-generation stay is worth noting too. Courts overseeing this litigation are fencing discovery to the specific model versions actually accused, rather than opening every current and future model to scrutiny in one sweep — a detail that should inform how data licensors think about scoping their own audit rights in contracts.