What did the New York Times v. OpenAI preservation order require OpenAI to retain and for how long?

1. What the order actually required: preserve and segregate “output log data”

Magistrate Judge Ona T. Wang’s May 13, 2025 preservation order directed OpenAI to “preserve and segregate all output log data that would otherwise be deleted on a going forward basis,” a phrase the court used to describe the output logs the company had been deleting under its ordinary retention and deletion practices ^{[1] [5]}. Legal analyses and reporting characterize that mandate to encompass ChatGPT consumer conversations and API conversation data — the records of user prompts and model outputs — that OpenAI would otherwise remove for privacy or retention-policy reasons ^{[2] [6]}.

2. How broadly the court framed retention — and the privacy collision

The order was written to capture output that would be deleted either because of user deletion requests or because of legal privacy obligations, meaning the court required preservation even where OpenAI’s normal policies or external privacy laws might have led to deletion ^{[1] [6]}. That sweeping scope transformed what began as a copyright discovery dispute into a broader question about mass retention of user-generated material and triggered immediate objections from OpenAI, which called the request overbroad and in tension with industry norms and its privacy commitments ^{[2] [7]}.

3. Temporal scope: “until further order” (i.e., indefinite) and what followed

The order’s timeframe was expressly open‑ended: OpenAI was directed to preserve covered output “until further order of the Court,” effectively creating an indefinite retention obligation until the court modified or terminated it ^{[1] [6]}. In subsequent litigation, judges considered objections and proportionality; by the summer months the preservation fight continued in district court proceedings, and later filings and a stipulation showed the parties moved to terminate OpenAI’s ongoing obligations under that preservation order ^{[2] [4]}. OpenAI has publicly stated that it was no longer under an order to retain consumer ChatGPT and API content indefinitely after September 26, 2025, and the parties filed a stipulation and order to terminate the obligation in October 2025 ^{[8] [3] [4]}.

4. Who was affected, exclusions, and the practical limits imposed

Reporting and court documents indicate the order potentially affected hundreds of millions of users’ consumer ChatGPT and API logs, though OpenAI later noted ChatGPT Enterprise was excluded from the preservation requirement; OpenAI also implemented restricted access protocols to limit who could see preserved material while the dispute continued ^{[2] [8] [9]}. The court and commentators debated proportionality and possible narrower alternatives (sampling, targeted searches), and the parties were encouraged to propose ways to limit retention’s scope — signaling the order’s practical implementation was contested and evolved after issuance ^{[10] [6]}.

5. Competing narratives and why the wording matters

Plaintiffs argued the logs were indispensable to test whether NYT content had been reproduced or used in training, while OpenAI framed the order as an unprecedented invasion of user privacy and an operational burden; both positions are reflected in the docket and company statements ^{[2] [7]}. Crucially, the order’s operative wording — preserve and segregate output log data “on a going forward basis until further order of the Court” — is what made retention indefinite in practice and why subsequent briefing, appeals, and a later stipulation were required to end or limit the requirement ^{[1] [10] [4]}.