On September 1, 2026, the U.S. Department of Justice (DOJ) filed a twenty-page "statement of interest" with the U.S. District Court for the Southern District of New York, marking the first time the federal government formally took a public position in the series of AI copyright lawsuits: using copyrighted text to train large language models constitutes "fair use" under copyright law and should not incur liability for infringement. The document was co-signed by Deputy Attorney General Stanley Woodward Jr. and others.
The reach of this document extends far beyond a single case. It not only addresses The New York Times' lawsuit against OpenAI and Microsoft filed in December 2023 — in which the Times accused the two companies of illegally using millions of copyrighted news articles to train AI systems and sought "billions of dollars" in damages — but also covers multiple consolidated lawsuits from book authors, publishers, and others being heard jointly in the same period. Judge Sidney Stein, who is presiding over the case, has required both parties to file motions for summary judgment by no later than September 4.
The Core Logic of the Legal Argument: Training and Output Phases Must Be Viewed Separately
The DOJ's core argument rests on a legal distinction: training activity and model output are two separate legal questions that cannot be conflated.
On the first fair-use factor, "the purpose and character of the use," the DOJ argues that the training process is "highly transformative" — rather than presenting the original articles to readers, the model converts all text into numerical representations, allowing it to learn statistical relationships among vocabulary, grammar, and knowledge, thereby acquiring capabilities such as text prediction, editing, translation, and content generation. The training act fundamentally changes the purpose of the original works, which no longer serve to convey information or provide entertainment to readers.
On the fourth factor, market harm, the DOJ's argument is equally pointed: the training act itself does not make protected content directly available to the public, so it cannot substitute for the original articles or harm their market value. Readers will not cancel their New York Times subscriptions because ChatGPT exists. As for the possibility that certain outputs may reproduce protected content verbatim — one of the core pieces of evidence cited by the Times in its complaint — the DOJ explicitly categorizes this as a separate legal question belonging to the "output phase," which cannot be used retroactively to establish that the training act itself was unlawful.
The DOJ also invokes a literary analogy in the document: Joan Didion, as a young woman, hand-copied Hemingway's short stories to understand their syntactic structure. Government lawyers argue that under the Times' litigation logic, every work Didion later published could have exposed her to legal liability — because her learning process and subsequent writing would be treated as a single chain of use. This analogy directly responds to the 2025 ruling in Kadrey v. Meta, which the DOJ explicitly criticizes in the document for its fourth-factor analysis being "fundamentally flawed."
The National Security Card: Elevating the Copyright Dispute into Geopolitical Competition
The DOJ's argument does not stop at purely technical legal analysis. The document explicitly introduces a national security framework: if copyright rules significantly raise the difficulty of developing large language models within the United States, it would harm the competitiveness of the U.S. AI industry and hand a competitive advantage to "foreign adversaries not bound by such constraints." The document also cites President Trump's executive order on "removing barriers to U.S. leadership in AI," characterizing AI dominance as a strategic issue bearing on "national security, prosperity, and economic mobility for all Americans."
The strategic intent of this argumentative framework is obvious: to recast the copyright dispute from a matter of private rights protection into a question of national competitiveness, thereby gaining the moral high ground. But critics have noted that this line of reasoning contains an internal tension — if restricting AI training would harm national security, then exactly whose interests stand in opposition to this "security threat"?
The Copyright Holders' Counterattack: Whose Side Is the Government On?
Graham James, a spokesperson for The New York Times, was blunt in his response: "The government is siding with a handful of trillion-dollar AI companies at the expense of countless American creators whose works have been taken." He further noted that AI and creators could have prospered together if AI companies simply paid fairly for content as copyright law requires — the government's position of allowing companies to take content without permission would undermine the very sustainability of human-created original content on which AI systems depend.
Steven Lieberman, lead attorney for the New York Daily News, raised questions at the constitutional level, arguing that the government's position "ignores the Copyright Clause of the U.S. Constitution — a provision crafted by the Founding Fathers to protect and incentivize the creation of original works," and noted that the DOJ's position contradicts views previously expressed by the U.S. Copyright Office.
This contradiction carries substantive significance. The Copyright Office sits within the Library of Congress and is theoretically independent of the executive branch; if the DOJ and the Copyright Office publicly diverge, it would create a rare internal rift within the legal system, forcing courts to choose between two government bodies.
The Undisclosed Detail
According to a July 2026 Financial Times report, OpenAI was at the time in discussions with the Trump administration over a plan to transfer a 5% equity stake to the federal government. If those negotiations were real, the DOJ had an undisclosed potential conflict of interest when it filed court documents supporting OpenAI. The legal commentary website Above the Law highlighted this omission in its September 3 analysis, suggesting that it casts suspicion of selective bias on the position taken in the entire document.
The DOJ filed a "statement of interest," not formal party status in the litigation. This procedural choice in itself means the government's position carries no binding legal force — the court may adopt it or disregard it. But in a case involving billions of dollars in damages that is widely regarded as the landmark copyright case of the AI era, the federal government's explicit alignment carries political and public-opinion weight that cannot be ignored.
What This Document Can and Cannot Change
The DOJ's statement has clear boundaries in terms of legal effect. The document explicitly states that its position applies only to the training phase, and does not cover the means by which training data is obtained, nor does it cover model outputs that may reproduce protected content verbatim. This means that even if the court fully adopts the government's position, OpenAI's legal exposure at the output level remains — the evidence cited in the Times' complaint of "ChatGPT reproducing NYT articles verbatim" will still undergo independent legal review.
From a broader perspective, this document marks the first time the U.S. federal government has formally taken a public position on AI copyright issues, filling a policy vacuum that had persisted for more than two years. Before this, AI companies had each invoked the fair use doctrine in court on their own, without systematic support at the government level. Now that gap has been filled — albeit in a manner that comes with considerable controversy.
For the global AI industry, the signaling value of this document already exceeds its legal binding force. The United States is currently the world's primary base for large language model development. The federal government's public placement of AI training under the fair use umbrella will provide a powerful reference point for other jurisdictions currently formulating AI copyright rules — whether it is the European Union's discussions on accompanying legislation for the AI Act, or the copyright exemption proposals being advanced in Japan and the United Kingdom, none will be able to avoid the influence of this American precedent.
Assessment
The DOJ's argument is not technically flawless: the "transformative" nature of the training process is real, but "scale" is the most vulnerable link in this logic. There is an essential difference in magnitude between a writer hand-copying Hemingway and a company valued at hundreds of billions of dollars systematically scraping tens of millions of works, and the current fair use framework does not provide clear legal standards for such differences in scale. How courts ultimately handle the "scale exception" question is where the real suspense of this litigation lies.
The political economy behind the document shows that when the government is both an industrial policy maker and a potential shareholder of leading AI companies, its proactive intervention in private copyright litigation on national security grounds is itself an exercise of power that warrants independent scrutiny. Controversy on this dimension will continue to intensify as the case moves into substantive trial proceedings.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接