What changed on September 1, 2026
The United States filed a Statement of Interest (Doc. 316) in In re OpenAI, Inc. Copyright Infringement Litigation, 25-md-3143 (S.D.N.Y.). The filing says the United States has a strong interest in AI competitiveness and national security and asks the court to reject the argument that training large language models on copyrighted texts is categorically infringing. It is the government’s position in active litigation, not a judicial holding.
The Associated Press report is useful secondary corroboration. The primary filing controls the description of what the government actually argued.
Keep the three copyright stages separate
| Stage | What the filing addresses | What remains open |
|---|---|---|
| Acquisition / storage | The developer collects and stores content before training. | Whether a particular acquisition, copy or storage practice was authorized or infringing. |
| Training | The DOJ focuses on copying works as learning material and argues the use is transformative and fair. | The court’s application of Section 107 to the record, including facts and competing evidence. |
| Outputs | The model responds to user queries after training. | Claims about memorization, substitution, market harm and output-specific conduct. |
Case timeline and posture
- December 2023: The New York Times sued OpenAI and Microsoft over alleged use of Times works in AI development and alleged output and market harms.
- April 4, 2025: Judge Sidney H. Stein largely denied the defendants’ motion to dismiss, leaving the claims to proceed.
- 2025–2026: Related publisher and author matters proceeded in consolidated MDL discovery and motion practice.
- September 1, 2026: The DOJ filed its Statement of Interest, urging a training-stage fair-use analysis and distinguishing other stages.
- Now: No court ruling adopting the DOJ’s position is established by this filing. The operative decision remains pending.
Bounded implications for legal and policy teams
For a publisher, preserve separate evidence about source acquisition, permissions, licensing, retention and observed outputs. For an AI developer, the filing does not remove the need to document dataset provenance, filtering, memorization controls, output safeguards or the commercial effect of a particular use. For counsel, avoid the shortcut “the DOJ says training is legal.” The narrower and supportable statement is that the DOJ advocates fair use for the training-stage copying question in this litigation.
The cross-case AI copyright training-data landscape remains the place for comparative case and policy coverage. This page owns the NYT v OpenAI docket, filing and stage-by-stage status.
What to record next
- Attach Doc. 316 and identify each proposition it supports.
- Map the exact dataset or content route to acquisition, storage, training and output records.
- Keep advocacy, judicial holdings, party allegations and vendor claims in separate fields.
- Record unresolved market-harm, memorization and output questions as Unknown until the docket supports a conclusion.
- Set a review trigger for the next court order, briefing deadline or discovery ruling.
FAQ
Did the DOJ filing decide NYT v OpenAI?
No. It is a Statement of Interest advocating the United States’ position. The court’s decision remains pending.
What fair-use question does the filing address?
It focuses on copying written works as learning material during the training stage. It distinguishes acquisition/storage and output-stage questions.
Does the filing make every AI training dataset lawful?
No. The filing does not decide whether particular acquisition, storage, licensing, memorization or output practices were authorized or infringing.
Sources checked
- DOJ Statement of Interest · Doc. 316, checked 2026-09-03.
- NYT v OpenAI case reporting, checked 2026-09-03.
- AI copyright training-data landscape, checked 2026-09-03.
Operational information, not legal advice. Verify current terms, account configuration and applicable professional duties before use.