In re OpenAI Copyright Infringement Litigation: the amicus briefs

What these filings are

The parties have cross-moved for summary judgment on fair use. The amicus briefs cite OpenAI’s and Microsoft’s motions (ECF 2461-1 at 7, citing Dkt. 1720, 1973, 1975 and 1980) and the plaintiffs’ briefs (ECF 2394-2 at 10, citing Dkt. 1982 and 2108). An amicus is not a party. Each of these briefs was filed as an exhibit to a motion asking the court’s permission to file it, and ChatGPT Is Eating the World reported on 2 November that all 51 remain pending the court’s approval. The court has limited proposed amicus briefs to 6,000 words (ECF 2404-1 at 16, citing ECF 1663).

The count has moved. On 31 October the same site gave an unofficial count of 52: 26 for plaintiffs, 18 for defendants and 8 for neither. That post’s web address says 54. Its 2 November post and its compiled list give 51, with 7 for neither. We have not checked the full docket and use the list of 51 below.

A note about our source. The author of ChatGPT Is Eating the World says in the 31 October post that they are among the amici (“myself included”). Their list does not say which brief is theirs. We rely on the site only for the list, the links and the counts. The side labels are the site’s.

All 51 briefs

Each name links to the RECAP copy of the brief on CourtListener. Names and side labels are as given on ChatGPT Is Eating the World’s list. “Read” marks the 13 briefs summarized on this page.

Four entries look different from the rest, and we note them as they appear without explaining them. Two are dated well before the late-October wave: the AI Rights Project (ECF 2099, 14 October) and Kevin Kelly (ECF 2112, 16 October). WME’s entry links to a RECAP file numbered 1826, out of sequence with its neighbours, while its docket link points to entry 2456. Michael Moritz (ECF 2434) has no PDF on the list. For the AI Coalition for Data Integrity (2386) and Kevin Kelly (2112), the docket links point to entries 2385 and 2111.

For the plaintiffs (26)

For the defendants (18)

For neither side (7)

What the plaintiffs’ amici argue

We read five of the 26. Page numbers are the ECF page stamps. Each position is the amicus’s argument, not a finding.

What the defendants’ amici argue

We read five of the 18.

Neither side

We read three of the 7.

Recurring themes

• Purpose and transformativeness. Plaintiffs’ amici frame the use narrowly: training on these works to produce outputs of the same kind (ECF 2393-1 at 8; ECF 2420-2 at 7). Defendants’ amici frame it as building a general-purpose tool (ECF 2429-1 at 8; ECF 2447-1 at 9). Both sides cite Warhol.
• Market dilution. The most contested point. Plaintiffs’ amici say harm from non-infringing outputs counts (ECF 2393-1 at 21-23). Defendants’ amici call it unprecedented and constitutionally suspect (ECF 2394-2 at 10; ECF 2429-1 at 19-23). The neutral academic briefs say it is unproven and should not be inferred (ECF 2380-1 at 8; ECF 2308-1 at 20-21).
• Licensing market. Plaintiffs’ amici list deals (ECF 2373-1 at 21-22; ECF 2393-1 at 18-20; ECF 2460-1 at 13-15). The other side calls a market for a transformative use circular or infeasible at scale (ECF 2429-1 at 23-24; ECF 2461-1 at 15; ECF 2308-1 at 18-19). Gans asks when the market existed and what each deal actually licensed (ECF 2381-1 at 19-22).
• Memorization, outputs and training. Plaintiffs’ amici point to verbatim output in the record (ECF 2373-1 at 18). Defendants’ amici and Sag treat memorization as limited and say outputs under realistic conditions are what matter (ECF 2447-1 at 13-14; ECF 2308-1 at 12-14).
• The Third Circuit’s ROSS decision. Cited by both sides. Plaintiffs’ amici rely on its rejection of “intermediate” copying to build a competitor (ECF 2373-1 at 8; ECF 2393-1 at 21). Others stress that it distinguished generative models that can produce original expression (ECF 2407-1 at 16; ECF 2308-1 at 22).
• How the data was acquired. EFF says acquisition does not matter if the use is fair (ECF 2407-1 at 10). Sag says shadow-library sourcing belongs under the fourth factor, not as a separate use (ECF 2308-1 at 15-16, 23).

The United States’ Statement of Interest

The United States filed a Statement of Interest on 1 September 2026 (ECF 1682 in the MDL; the copy we read is stamped ECF 316 in No. 1:25-cv-03483). It argues that training on written works is “exceedingly transformative” (at 11). It says required licensing fees would favour the largest technology companies (at 5). It calls the dilution reasoning in Kadrey v. Meta “deeply flawed,” because outputs that are not substantially similar cannot cause cognizable market harm (at 16-17). Its example is Joan Didion, who typed out Hemingway’s stories to learn how his sentences worked (at 17).

At least six of the briefs we read take it on. The members of Congress answer it directly (ECF 2404-1). The copyright professors and the music groups reject the Didion comparison (ECF 2420-2 at 13; ECF 2393-1 at 17). The News/Media Alliance and the AI startups dispute its claim that only the largest companies can afford licenses (ECF 2373-1 at 22-23; ECF 2460-1 at 15). Samuelson and Sprigman note its criticism of the Copyright Office’s report on dilution (ECF 2380-1 at 10 n.1).