# Case Facts: Kadrey v. Meta Platforms Inc.

## Case Information
- **Case No.:** 3:23-cv-03417-VC
- **Court:** United States District Court, Northern District of California
- **Judge:** Hon. Vince Chhabria
- **Filed:** July 2023
- **Motion to Dismiss Ruling:** November 25, 2024

## Plaintiffs
- Richard Kadrey (author of "Sandman Slim" series)
- Sarah Silverman (comedian/author of "The Bedwetter")
- Christopher Golden (horror/fantasy author)

## What Meta Actually Did (According to Complaint)

### 1. Acquisition of Books
- Downloaded copyrighted books from **Library Genesis (LibGen)**
- Used **BitTorrent** to obtain books from piracy networks
- Accessed **Anna's Archive** for copyrighted works
- Obtained datasets like **"Books3"** (containing 170,000+ pirated books)
- All without author permission or payment

### 2. The "Annotation" Process
- Hired contractors to read the books
- Claimed purpose was to "annotate" them for training data
- **No evidence provided that meaningful annotation occurred**
- Contractors essentially just read the books on Meta's dime
- Books were kept in Meta's systems

### 3. Training Decisions
- Used many books to train LLaMA (Large Language Model Meta AI)
- **Decided NOT to train on some books**
- **Kept all the books anyway** (did not delete)
- No compensation to authors regardless of use/non-use

### 4. The Defense
Meta argued:
- The copying was "transformative" because LLaMA doesn't reproduce books verbatim
- Authors suffered no "cognizable harm" because outputs are different
- Using books as training data isn't infringement if the model doesn't memorize them
- Downloading from piracy sites doesn't constitute DMCA circumvention

## Judge Chhabria's Ruling

### Claims Dismissed
1. **DMCA Circumvention Claims** - Court found no technological protection measures were circumvented
2. **Certain Copyright Claims** - Court skeptical that training data use alone constitutes infringement

### Key Quotes from Ruling
> "The plaintiffs have not adequately alleged that Meta's alleged copying of their books constitutes copyright infringement."

> "The outputs of LLaMA do not contain substantial portions of the copyrighted works themselves."

> "The Court is skeptical that the mere use of copyrighted material as training data, without more, constitutes actionable infringement when the resulting AI model does not reproduce or display the original works."

### Claims Allowed to Proceed
- Direct copyright infringement claims (for the initial copying)
- Unjust enrichment
- Negligence (for data security)

## The Absurdity

Meta's position essentially says:
1. ✅ We can download your book from piracy sites
2. ✅ We can have employees read your entire book
3. ✅ We can keep it on our servers indefinitely
4. ✅ We can decide whether or not to use it for training
5. ✅ We never have to delete it
6. ✅ We never have to pay you
7. ✅ And it's all legal because our AI doesn't quote you verbatim

**If this logic holds, what's to stop anyone from:**
- Creating a "reading club" where they torrent books?
- Keeping massive libraries of pirated content "for potential AI training"?
- Never paying authors because "we might not train on it"?
- Claiming everything is "research" or "annotation"?

## Related Cases

### Silverman v. OpenAI
- Similar allegations against OpenAI's GPT models
- Books3 dataset (pirated books from Bibliotik tracker)
- Some claims dismissed, others proceeding

### Andersen v. Stability AI
- Visual artists suing Stability AI (Stable Diffusion)
- Training on copyrighted images without permission
- Similar "transformative use" defense

### New York Times v. OpenAI
- Major news publisher suing for training on articles
- Different because NYT can show verbatim reproduction
- Proceeding to discovery

## Why This Matters

If Meta's defense succeeds:
- Copyright becomes optional for AI training
- Authors lose control over their work
- "Transformative use" becomes infinite loophole
- Only remedy is if AI accidentally quotes you verbatim
- Big Tech can legally build on stolen content

## The Counter-Argument

Authors argue:
- **Initial copying is infringement** (regardless of training outcome)
- **Making derivative works is infringement** (model weights contain compressed book knowledge)
- **Commercial use is not fair use** (Meta profits billions from LLaMA)
- **Market harm is real** (who buys books when AI can summarize them?)
- **The law should protect creators**, not just tech companies

## Current Status (as of Feb 2026)

- Case proceeding to discovery on remaining claims
- Meta must produce documents about training process
- Authors seeking class action certification
- Similar cases being consolidated for efficiency
- **No final ruling on the core copyright questions yet**

## Sources

- Court dockets from PACER
- Electronic Frontier Foundation coverage
- The Hollywood Reporter legal analysis
- Stanford Law School AI + Copyright commentary
- Public filings in Kadrey v. Meta

---

## The Satirical Question

**If Meta can do this, can anyone?**

That's what this website explores. If the answer is "no, only Meta can," then we have a problem with our legal system. If the answer is "yes, anyone can," then copyright law is dead.

Either way: **absurd.**