When the Supreme Court reversed a billion-dollar copyright verdict against Cox Communications in March, the message seemed clear: merely knowing your service is being used to infringe copyrights isn't enough to make you liable.

Three months later, courts are beginning to fill in what "more" actually means. A handful of district courts have now applied Cox to copyright disputes, one involving NVIDIA and another X (formerly Twitter). Although decided on Rule 12 motions (where allegations must be taken as true), they tell us how courts are developing a key post-Cox framework.

The Cox Decision

The Supreme Court's decision in Cox Communications, Inc. v. Sony Music Entertainment, 607 U.S. ___ (2026) held that contributory copyright liability requires intent, and intent can be shown in only two ways: the defendant either actively induced infringement through specific acts, or provided a service with no substantial non-infringing uses. Cox, an internet service provider that had received over 163,000 infringement notices and terminated only a handful of subscribers, nevertheless escaped liability because it hadn't done either. Providing internet access is not a service tailored to infringement. And discouraging subscribers from pirating — even halfheartedly — is not inducement.

What Cox did not do (as Justice Sotomayor pointedly noted in her concurrence) was draw a bright line around every technology platform. The majority's framework still leaves room for defendants whose conduct goes beyond passive provision of a service. Three recent decisions show courts working out where that room begins and ends — at least at the pleading stage.

A Platform With a Purpose-Built Tool

In Nazemian v. NVIDIA Corporation, No. 24-cv-01454-JST (N.D. Cal. May 5, 2026), authors alleged that NVIDIA trained large language models on copyrighted books scraped from shadow libraries, and that NVIDIA provided customers with scripts specifically designed to download and preprocess a dataset called The Pile — which contained those books. NVIDIA argued its “NeMo Megatron Framework” was a general-purpose AI development platform, just like Cox's internet service. 

To understand what's at stake, some context helps. NVIDIA's NeMo Megatron Framework is a software toolkit that companies use to build and train AI language models (LLMs). Within that toolkit, NVIDIA allegedly included scripts, small automated programs, that would locate and download The Pile from the internet and prepare it for use as training data. It's that specific functionality — not the broader toolkit — that ended up at the center of the case.

The court denied the motion to dismiss on contributory infringement — but the ruling is carefully limited to what plaintiffs alleged, not what they proved. NVIDIA argued that the NeMo Megatron Framework as a whole was a general-purpose AI development platform with many legitimate uses, just like Cox's internet service. But Plaintiffs didn't attack the whole framework. They targeted those specific scripts that NVIDIA allegedly built and distributed for the sole purpose of pulling down a dataset of pirated books. Accepting those allegations as true, the court found the scripts plausibly had no non-infringing use — exactly the "service tailored to infringement" theory Cox preserved. Whether NVIDIA actually designed those scripts with infringing intent, and whether The Pile actually contained plaintiffs' works, remains to be established through discovery.

The preliminary lesson from Nazemian: claims against purpose-built tools that plausibly have no legitimate use may survive Cox at the pleading stage. Whether that theory holds up on a developed factual record is a key question worth watching.

Specific Notice, Specific Inaction

Michael Grecco Productions, Inc. v. Twitter, Inc., No. 8:24-cv-04878-MEMF-JC, 2026 WL 917606 (C.D. Cal. Mar. 31, 2026), offered a different path to surviving a Rule 12 motion. A photographer whose images appeared on X (formerly Twitter) had sent over 100 DMCA takedown notices. Twitter did nothing. The court distinguished Cox in a footnote: Cox had at least attempted to discourage infringement; Twitter had received notice of specific infringing works and — taking plaintiff's allegations as true — simply ignored them. Refusing to act after concrete notice is not passive provision of a service. All three copyright theories survived.

What Grecco does establish, at least provisionally, is that Cox doesn't protect willful blindness to specific infringement. A platform that can point to an active compliance program, as Cox could, is in a different position than one alleged to have done nothing at all.

Preliminary Takeaways

Read together, these two decisions sketch a preliminary post-Cox framework: Defendants who receive concrete notice of specific infringing works and fail to act may still face viable claims, at least at the pleading stage. Conversely, Defendants who operate general platforms without that notice may largely be protected. And significantly, Defendants who allegedly build tools specifically designed to facilitate access to infringing material may face the most exposure.