Skip to content

Unsealed Briefs Authors Case v Microsoft/OpenAI Guide

Unsealed briefs in Authors' Case v. Microsoft/OpenAI just dropped. Here's how non-lawyers pull the docs, spot key admissions, and use them right now.

7 min readBeginner

Primary docs beat another quote thread

Mid-September 2026, public versions of the big summary-judgment papers in the authors’ fight against OpenAI and Microsoft landed on the docket. Headlines grabbed the same Slack lines. Almost nobody handed you the PDFs, the pin cites, or a way to check redactions yourself.

If you ship product on LLMs, write for a living, or draft internal AI rules, secondary recaps are a weak diet. You want the memorandum and the Rule 56.1 statement – the numbered fact list that ties each admission to an exhibit – not a free-floating screenshot.

Reading a sealed-then-unsealed filing feels a bit like being handed a heavily marked exam script: the answers that survived are sharp, the black bars tell you what still matters to the parties, and the grading isn’t done. That gap between “viral quote” and “what the PDF actually supports” is the whole reason to open the file.

What most write-ups skip

Recaps and HN threads (hundreds of comments landed fast after the Authors Guild post) chase atmosphere. They rarely give page or SUF numbers, rarely walk Microsoft’s alleged vicarious-liability theory, and almost never show how plaintiffs line torrenting, training copies, and cross-company transfers against the four fair-use factors.

Friction is real. Public re-files still omit swaths of exhibits. CourtListener’s RECAP mirror is free only after someone donates a document; brand-new attachments can sit on the “Buy on PACER” button at $0.10/page for a while. Miss the Corrected Rule 56.1 SUF and you miss the dense spine of the motion.

Non-lawyer workflow: two PDFs, five searches

Grab the cleaned public set the Authors Guild hosted the week of the re-file. That’s the Class Plaintiffs’ Memorandum of Law (about Dkt. 1982) and the Corrected Rule 56.1 Statement of Undisputed Material Facts (Dkt. 1987) in In Re: OpenAI, Inc. Copyright Infringement Litigation, 1:25-md-03143 (S.D.N.Y., Judge Sidney H. Stein; Mag. Ona T. Wang). Original Authors Guild case path: 1:23-cv-08292.

  1. Open the Authors Guild release and download both linked PDFs. No login.
  2. Memo first (narrative + fair-use argument). SUF second (numbered facts + exhibit pins).
  3. PDF search: LibGen, Project Clear, Jack Clark, Gogineni, Gates, substitut. Write down paragraph or SUF numbers.
  4. Cross-check stamps on CourtListener. Filter September 2026 if you need related opposition papers later.

Day-one bar: you can cite a pin instead of a viral sentence.

Two-column note works better than another tab full of commentary – left: quote + SUF/memo pin; right: training data / market harm / willfulness.

Edge realities before you over-claim

The public “unsealed” memo and SUF are not the full sealed originals. Per the sealing setup and the Guild’s hosted files, contested and highly confidential passages stayed out; many exhibits remain sealed or attorneys’-eyes-only on the docket. Always match the PDF in your hand to the latest ECF entry.

Project Clear privilege fights cut both ways. Magistrate Wang ordered production of a large set of non-privileged Slack messages; a later district-court opinion set aside parts of the waiver ruling on certain lawyer communications. Result: the SUF quotes public slices of the deletion discussion, not every internal thread.

RECAP lag is ordinary. High-profile dockets fill in fast, but a fresh attachment can still bill on PACER until a user uploads it. For these two briefs, the Guild PDFs are the immediate free route as of the September 2026 release.

Hearing timing after oppositions and replies is still fuzzy in the public materials – described as early 2027, not a firm calendar day.

What the public record supports (AI-user cut)

More than four million books via LibGen torrents, plus Books3 and other sources – that’s the scale plaintiffs put in the public memo/SUF. Microsoft’s awareness of LibGen shows up as early as an April 2019 walkthrough of early GPT-3 work for Bill Gates and then-CTO Kevin Scott, according to those same papers.

May 2020, Jack Clark (policy role at the time) on stronger GPT-class models: genre-fiction authors would worry about Amazon substitution; the work “will make people unemployed” (SUF ¶749 area as cited in the Guild summary). 2022 hire Tarun Gogineni’s stated research interest included models finishing the last two A Song of Ice and Fire books; author blowback on datasets framed as “acceptable economic disruption,” with a bleaker riff on machines making slop for machines.

Summer 2022: Project Clear. Slack started life as “excise-libgen,” later renamed under then-GC Jason Kwon. Public portions describe scrubbing LibGen files and references when scrutiny risk rose – McGrew on timing to “excise Libgen.” Optics (including fear of a bad Hacker News hit) sit next to legal risk in the story plaintiffs tell.

Same-week news-publisher papers, as covered by Bloomberg Law (Sept 17-18 2026), add Nadella on pirate downloads as “absolutely” illegal and product-side language that tools are “largely substitutive, period.”

Plaintiffs’ theory: pirate acquisition, training copies, and transfers among OpenAI, Microsoft, and contractors fail fair use across the factors – commercial purpose, amount, market substitution/licensing harm – and Microsoft faces vicarious liability. Defendants get their oppositions; nothing here is a final judgment.

When executives themselves lean on “substitutive,” what does that do to the old casual line that chatbots only “help writers brainstorm”? Sit with that before you paste another undifferentiated summary into a policy doc.

Ten-minute trace (Clark line)

Want the “unemployed” sentence in context? Memo PDF → intro / LLM-as-replacement stretch → jump to the Guild-cited SUF paragraph. You get surrounding policy discussion and the exhibit pin, not a detached quip.

Same pass for Gates/LibGen or Project Clear. Pin in hand, you can drop an accurate cite into an AI usage policy, a creator newsletter, or a short note on why some models balk at full-book continuation prompts.

While the MDL is live

  • Train or fine-tune? Provenance memos now. “We scraped the open web” lands softer once shadow libraries are on the public record.
  • Write or publish? Market-harm sections plus internal “substitutive” wording are use in licensing or opt-out talks.
  • Daily ChatGPT user? Training fights feed refusal behavior, filters, and future licensed corpora – what the model will cheerfully emit can shift.

Watch 1:25-md-03143 for defense briefs and any further unsealing orders. Redactions still hide a lot of exhibits. Trial/hearing language in the public Guild materials stays at “early 2027.”

FAQ

Where do I get the free unsealed briefs without paying PACER?

Authors Guild page above – public memo + SUF PDFs, no account. CourtListener RECAP for 1:25-md-03143 is free after donation. Pay PACER’s $0.10/page only when the buy button is still there.

Are the shocking quotes sealed?

The Clark, LibGen-knowledge, Project Clear, and Gogineni passages plaintiffs lean on appear in the Guild’s public PDFs. Underneath, many exhibits and confidential business chunks stay sealed or AEO. Privilege skirmishes also mean not every Project Clear Slack line made it out. Compare your file to the latest docket stamp; sealing fights aren’t over.

Does this mean ChatGPT training was illegal?

No ruling yet. Plaintiffs asked for partial summary judgment: pirate downloads, training copies, and distributions weren’t fair use, and Microsoft should face vicarious liability. OpenAI and Microsoft will argue important fair use and file oppositions. The unsealed record hardens the knowledge/willfulness narrative. It does not end the case. Public scheduling talk points to a hearing in early 2027, without a locked date in the Guild release. Treat the PDFs as strong evidence – not a final scoreboard.

Download both PDFs, run the five searches, and write one short note on how the pins change your own AI workflow. That’s the concrete next step.