what sign up
nixm nixm

← learn

NYT v. OpenAI — how a copyright case became about your chats

August 31, 2026

What began as a copyright dispute became the defining AI-privacy story: courts ordered OpenAI to preserve even deleted chats, then to produce 20 million user conversations to opposing lawyers — affirmed in January 2026. In July 2026 the plaintiffs moved for sanctions, alleging logs were deleted anyway. The lesson for users is simple: stored chats are evidence. nixm stores none.

How did a copyright case reach my chats?

The New York Times sued OpenAI in late 2023 over training on its articles. To prove ChatGPT reproduced their content, the news plaintiffs wanted to see what ChatGPT actually output to real users — which meant the logs. In May 2025, Magistrate Judge Ona T. Wang ordered OpenAI to preserve all output logs "that would otherwise be deleted," including user-deleted chats and Temporary Chats. Four hundred million people's delete buttons, suspended by a case they'd never heard of. The matter now runs as a consolidated multidistrict proceeding in the Southern District of New York, combining a set of publisher and author suits.

What happened next?

OpenAI fought and lost, loudly — its COO called the demand baseless, Sam Altman called for an "AI privilege" akin to doctor-patient confidentiality, and users who tried to intervene were denied as non-parties. The blanket preservation was lifted in October 2025 (chats after September 26 returned to the normal 30-day deletion cycle) — but everything already saved stayed accessible to the litigation, and specific flagged accounts remained on hold. Reporting on the modified order notes it also excluded data originating in the European Economic Area, Switzerland and the United Kingdom.

Then came the twenty million

In November 2025, the court ordered OpenAI to produce a de-identified random sample of 20 million consumer conversations — a number OpenAI itself had once proposed as "surely more than enough" before reversing. The court noted the sample was a small fraction of the tens of billions of consumer logs OpenAI retains, and found user privacy adequately protected by de-identification and attorneys-eyes-only handling. District Judge Sidney Stein affirmed it on January 5, 2026 — and affirmed the whole sample, not the narrower set of conversations OpenAI had wanted to hand over instead.

The sanctions fight

This is where the story turned from privacy to conduct. OpenAI's December production was so heavily redacted the court deemed it unusable. In January 2026, counsel for the news plaintiffs told the court that more than a million of the logs OpenAI had been ordered to produce were substituted with other conversations, apparently because the originally selected ones had been deleted.

On July 9, 2026, the publisher plaintiffs moved for sanctions. Per reporting on the filing, the motion rests on two claims: that OpenAI told the court for more than two years that it could not search its training data and output logs for the plaintiffs' works when it had already built that capability, and that deletion of conversation logs continued despite the preservation order. Reporting on April 2026 testimony describes an OpenAI data privacy engineer stating the company had built a de-identified conversation dataset of roughly 78 million records and was using it to assess the scale of its own alleged infringement.

These are allegations in a live motion, not findings. Whether a court agrees is not settled. But the shape of the argument is worth noticing on its own: the dispute is no longer about whether logs exist. It is about how many, who could search them, and what happened to the ones that were supposed to be handed over.

What should a normal user take from this?

Three durable facts. Stored conversations are evidence, compellable like email. Deletion is a schedule with exceptions, and legal hold is the exception. And the only data that survived the whole saga untouched was data that was never logged — OpenAI's zero-retention API tier was carved out precisely because nothing existed to preserve. How long ChatGPT actually keeps chats.

Where does nixm stand in this?

On the never-logged side of the line, for everyone. The conversation is never written to storage — there is no log pile for any future case to sample. What persists is a distilled signal (intent, short summary, theme) that dissolves in seven days. Not a policy promise that yields to court orders — an absence that has nothing to yield.

Think outside the record

Open nixm and seed a signal — no account, and nothing you type joins anyone's tens of billions.

frequently asked

Why did a court preserve deleted ChatGPT chats?

News plaintiffs argued users skirting paywalls would likely delete those chats — so the court ordered even deleted and Temporary chats preserved as potential evidence.

Is OpenAI still preserving everyone's deleted chats?

No. The blanket order was lifted in October 2025 — but logs saved under it remain accessible to the case, and flagged accounts stay on hold.

What are the 20 million produced conversations?

A court-ordered, de-identified random sample of consumer chats given to plaintiffs' lawyers — affirmed in full in January 2026, drawn from tens of billions of retained logs.

What is the 2026 sanctions motion about?

Publisher plaintiffs moved for sanctions on July 9, 2026, alleging OpenAI wrongly told the court it could not search its logs, and that log deletion continued despite the preservation order. These are allegations, not findings.

Were users' names in the produced logs?

The sample was de-identified under a protective order with attorneys-eyes-only handling — the court held that adequately protected privacy; users' objections were not heard, as non-parties.

Was UK or EU data covered by the preservation order?

Reporting on the modified order says data originating in the European Economic Area, Switzerland and the United Kingdom was excluded.

What was exempt from all of this?

Zero-data-retention endpoints — nothing was logged, so nothing could be preserved or produced. The same principle nixm applies to every conversation.

Can I try nixm without an account?

Yes — open nixm.ai and seed a signal directly; no sign-up, and nothing you type is stored.

try it now — no account, nothing stored. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+