In a Third Amended Complaint filed June 25, the Times targets the custom built Microsoft training system behind OpenAI's chatbots, recasting an AI copyright lawsuit as a fight over the infrastructure that built the model.
The New York Times is no longer just suing the AI company. In a Third Amended Complaint filed on June 25, the newspaper reframed Microsoft's custom-built training supercomputer as the alleged instrument of copyright infringement, accusing Microsoft and OpenAI of engineering infrastructure whose purpose was to ingest and mimic licensed journalism without permission (Ars Technica). With Microsoft's market capitalization near $3 trillion and OpenAI's private valuation resting on the durability of its training pipeline, both defendants now sit inside a redefined legal perimeter.
The complaint pivots a roughly three-year-old case onto fresh legal ground. When the Times first sued OpenAI in late 2023, it focused on what the chatbot did: regurgitate articles nearly verbatim, sidestep paywalls, displace subscriptions, and falsely attribute claims to its reporting (NYT newsroom coverage). The amended complaint keeps those claims but adds a new one: that the custom-built Microsoft training cluster behind ChatGPT-style models was engineered to ingest copyrighted journalism without permission and to weight the highest-quality work more heavily inside the training run (Third Amended Complaint).
That distinction matters beyond courtroom choreography. Contributory copyright infringement, the legal theory the Times is now leaning on, treats the party that supplied the means of infringement as liable alongside the direct infringer. The earlier framing described Microsoft's hardware as generic cloud capacity OpenAI rented; the new framing calls it "unusually complex" infrastructure designed around the goal of mimicking licensed work (Ars Technica). If a court accepts that reframing, the result is not just another AI training lawsuit. It is a precedent that a cloud provider can be on the hook for designing a system whose purpose was infringement, regardless of whose model eventually runs on it.
Bloomberg Law described the amendment as the Times "narrowing" the case to focus on Microsoft's conduct directly, a procedural move that drops parts of the original 2023 theory while concentrating the remaining fire on the infrastructure partner (Bloomberg Law). The Times's own newsroom framing leans on discovery evidence: shared ChatGPT sessions showing near-verbatim Times excerpts and prompts explicitly written to bypass the newspaper's paywall (NYT). Those data points do not establish that every ChatGPT user engaged in infringement, but they show what the Times intends to put in front of a jury.
Three limits belong on the page. First, these are still allegations; the Third Amended Complaint is a pleading, not a finding, and neither Microsoft nor OpenAI had issued a public response to the new claims as of the Times's filing. Second, the "unusually complex" and "tailor-made" characterizations come from the complaint itself, and Ars Technica's summary uses similar quoted language; a court could read the same hardware as a routine evolution of Microsoft's Azure AI fleet. Third, the contributory infringement theory is harder to win than it is to plead, particularly when the underlying training process is itself contested. Other plaintiffs, including the authors and publishers suing OpenAI in parallel matters, will be watching the docket closely (CourtListener docket).
The next beat to watch is the response brief. Microsoft's first chance to rebut the new theory on the merits will arrive when it answers the Third Amended Complaint, and the choice between fighting the reclassification as a procedural defect and engaging on contributory infringement substance will signal whether Redmond plans to treat this as a winnable case or a settlement conversation.