DOJ files under § 517 on 1 Sep

On 1 September 2026, the United States filed a Statement of Interest in the Southern District of New York in In re: OpenAI, Inc. Copyright Infringement Litigation, MDL No. 25-md-3143 (SHS) (OTW). The ECF stamp on the RECAP copy also shows Case 1:25-cv-03483-SHS-OTW, Document 316, Filed 09/01/26, 20 pages. The same text appears on the MDL docket as Document 1682 in a Copyright Alliance-hosted mirror.

The filing opens: "The United States appears pursuant to 28 U.S.C. § 517 to inform the Court of its interest in this litigation." Signatories are Stanley E. Woodward, Jr., Associate Attorney General; Brett Shumate, Assistant Attorney General, Civil Division; and Michael Weisbuch, Senior Counsel to the Associate Attorney General. The document is dated September 1, 2026.

The United States says it has a "strong interest in the question whether training AI models on written works constitutes 'fair use' under copyright law." It states a strong interest in the Court rejecting "any argument that training LLMs on copyrighted texts violates copyright law."

Training fair use and EO citations

Executive-order frame. The Statement quotes the 23 January 2025 Executive Order Removing Barriers to American Leadership in Artificial Intelligence: it is U.S. policy "to sustain and enhance America's global AI dominance in order to promote human flourishing, economic competitiveness, and national security." It also cites the 2 June 2026 Executive Order Promoting Advanced Artificial Intelligence Innovation and Security on encouraging AI innovation and accelerating responsible adoption.

National security passage. The United States argues that rules that make robust U.S. AI development "significantly more difficult" "threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered." It cites GAO language on AI for intelligence analysis, weapon systems, and battlefield recommendations.

National Policy Framework quote. The Statement cites the March 2026 National Policy Framework for Artificial Intelligence for the proposition that, consistent with promoting innovation and free expression, the "training of AI models on copyrighted material," in and of itself, "does not violate copyright laws." Footnote 11 says the United States refers to the New York Times and OpenAI for simplicity, and that its legal arguments "apply similarly to all parties in this litigation and the related cases, including book authors and publishers."

Scope of the ask. The United States focuses on the training stage: "whether the use of copyrighted works at the training stage (copying works to feed data into the model as learning material) constitutes fair use." It separates that question from output-stage uses. It argues training is "extraordinarily transformative," citing among other cases Bartz v. Anthropic, Kadrey v. Meta, Authors Guild v. Google, and Google v. Oracle. On the fourth fair-use factor, it says a copy used in training does not "serve as a substitute for the original." It calls the Kadrey court's market-dilution approach "deeply flawed" for having "improperly collapsed LLM training and LLM outputs into a single continuous use."

Competition claim in the brief. The United States warns that an "erroneous fair use ruling would hamper competition in the market for LLMs, because only the largest technology companies might have the capital necessary to pay licensing fees," and that such fees would "disproportionately benefit legacy media outlets." Footnote 13 says the United States "takes no position on whether a licensing regime would be financially or logistically feasible," and notes publishers can still license specialized real-time or paywalled access even if training is fair use.

Closing ask. The Statement concludes it would be "problematic" and "legally incorrect" to impose broad copyright liability that would generally render training of AI models impermissible without licensing, and that LLM training is "consistent with that creative 'progress' that is the basic constitutional objective of copyright itself."

Government enters the fair-use fight

A Statement of Interest under § 517 is not a party complaint and not a judgment. It puts the Department of Justice's legal position on the MDL record in the OpenAI copyright consolidated litigation that includes the New York Times action and related publisher and author cases.

For operators, the dated change is institutional: the United States has now told the S.D.N.Y. MDL court, in writing, that it wants training-stage fair use preserved as a matter of copyright doctrine tied to AI competitiveness and national security executive orders. Publisher plaintiffs in the same MDL still press infringement theories. Those tracks remain open. This filing does not dismiss them.

What counsel should brief Monday

  • Read the primary PDF. Archive Document 316 / MDL Document 1682. Do not rely on secondary paraphrases for the EO citations or the National Policy Framework quote.
  • Separate training from output. The United States expressly limits its fair-use focus to training-stage copying. Output regurgitation claims stay live as a distinct issue in the same MDL.
  • Treat this as advocacy. § 517 puts the government's interest before the court. It does not decide fair use.
  • Map licensing exposure. If your stack trains or fine-tunes on news text, brief boards on how this Statement frames compulsory licensing as a competition and national-security risk, while still allowing voluntary specialized licenses.
  • Track related publisher suits. Parallel complaints (including midsize publishers) continue to allege training and product harm. Calendar the MDL captions 25-md-3143 and 1:25-cv-03483.
  • Do / buy / ignore / distrust. Do: update the litigation watch memo with the 1 Sep 2026 filing date and the training-only scope. Buy: outside copyright counsel time to reconcile this Statement against your training-data provenance log. Ignore: social claims that the DOJ "ended" publisher copyright suits. Distrust: any vendor slide that treats this Statement as a license grant or safe harbor.

What this Statement does not settle

  • That training on copyrighted news is fair use as a final judicial holding. The Statement argues that position; the court has not so held in this filing.
  • That output-stage copying or substitution claims fail. The United States says those may raise different questions.
  • That every publisher licensing market is legally irrelevant. The brief distinguishes voluntary specialized licenses from a rule that training itself requires a license.
  • That national-security executive orders rewrite § 107. They are cited as interest and policy context. They are not a statutory amendment.
  • Party discovery facts about what OpenAI actually copied. The Statement relies on the litigation's described training pipeline and case law; it is not a forensic inventory.
  • X or press claims that OpenAI or Microsoft funded any particular plaintiff newspaper. That topic is outside this Statement.

Gaps after Document 316

  • The court's next order responding to the Statement of Interest, if any.
  • Full briefing schedules and reply papers from the New York Times and other MDL plaintiffs on this filing.
  • Defendant responses that adopt or distinguish the United States' training-only fair-use frame.
  • Whether Congress acts on the National Policy Framework's licensing-framework recommendations referenced in the Statement.
  • Consolidated treatment of parallel midsize-publisher complaints filed after this Statement.
  • Any trial or summary-judgment ruling that actually decides training-stage fair use in this MDL.

Monday action: Policy and IP counsel should file the 1 Sep 2026 Statement under the OpenAI MDL watch list, brief leadership that DOJ now argues training-stage fair use as U.S. interest tied to AI dominance executive orders, and keep publisher suits open as unresolved. Do not treat the Statement as a judgment or a product license.