- Anthropic is now facing a class-action copyright trial on Dec. 1, 2025.
- The total number of works in the class will be determined by a list submitted to the court on Sept. 1. But it has to be some number below 7 million (the total number of files downloaded).
- If a jury in San Francisco finds Anthropic engaged in willful (knowing it’s illegal) infringement by downloading copies from 2 shadow libraries (Library Genesis and PiLiMi), Anthropic’s chances of facing business-ending damages award in the billions increases.
- The basic reason is the multiplier effect from the figure for statutory damages (picked from a range of $750 – $30,000 or up to $150,000 if jury finds willful infringement) x total number of works infringed (a number to be determined but something under 7 million).
- It’s hard to believe Anthropic may face a similar fate as Napster, the notorious file-sharing software provider that was prominent in illegal music file sharing that declared bankruptcy.
- But it is possible, if not probable. We ran the numbers.
More and more it looks like December 1, 2025 will be a day of reckoning for Anthropic.
After Judge Alsup certified a class action of authors’ whose books were downloaded by Anthropic in 2 shadow libraries of pirated books (Library Genesis and PiLiMi), and after the judge ruled that such downloading and storage by Anthropic in a library was NOT fair use, Anthropic faces the herculean task of convincing a jury in San Francisco that (i) its conduct was not willful and (ii) the jury should pick a dollar figure on the low end of the statutory range, such as $750 per work.
At trial, Anthropic is at the mercy of the jury.
Our prior analysis here and here showed that Anthropic can easily face business-ending liability (or at least a judgment forcing bankruptcy or the need for an infusion of capital from outside investors to help cover the bond in the amount of the damages that Anthropic will have to post during its appeal).
Our dire assessment is supported by Anthropic’s immediate hiring of 2 new law firms with prominent trial attorneys to assist, if not lead, the defense with the other law firms already representing Anthropic in the two copyright suits against it. This is a code red situation.
probabilities of various statutory damages awards v. anthropic
Our assessment is further confirmed by estimation of probabilities of different damages awards ChatGPT generated using the Monte Carlo simulation. We used this technique given yet-to-be-determined variables (number of works and dollar amount within the statutory range), so we could get a better sense of the business risk Anthropic faces.
Under our simulation, we asked ChatGPT not to favor any finding of willfulness by Anthropic or to include the possibility that the court would override the jury’s amount using remittitur. In other words, these probabilities relate to the jury’s calculation and excludes remittitur as a backstop. Accordingly, ChatGPT assumed the jury’s finding of willfulness was a 50/50 coin flip. (We also assume ChatGPT is properly applying the Monte Carlo method and is not biased against Anthropic.)
Here’s how ChatGPT estimated the probabilities of different awards using Monte Carlo simulation:
| Potential Statutory Damages Awards v. Anthropic | Probability |
| <$1 billion | 4.69 |
| $1 billion to $5 billion | 16.9 |
| $5 billion to $10 billion | 14.12 |
| $10 billion to $25 billion | 25.55 |
| $25 billion to $50 billion | 16.44 |
| $50 billion to $100 billion | 12.61 |
| $100 billion or more | 9.68 |
As you can see the greatest probability right now is 25.55% for an award between $10 billion to $25 billion.
Shockingly, the mean expected award was ~$34.1 billion, according to ChatGPT.
Here’s a histogram of the Monte Carlo simulation for possible damages awards:
We can depict these probabilities in a pie chart below. As you can see, the biggest blue slice is the bottom right, the range for $10 billion to $25 billion dollars in statutory damages. And the smallest pie slice is for the lowest award under $1 billion.
A simpler categorization of the ranges lumped together:
- Mean expected award: ~$34.1 billion
- Probability > $1B: 95.31%
- Probability > $10B: 64.28%
- Probability > $50B: 22.29%
- Probability ≥ $100B: 9.68%
Shockingly, the probability of a $100 billion or more award is nearly double the probability of an award less than $1 billion.
why is the probability of an award under $1 billion so low????
I asked ChatGPT why the probability of an award under $1 billion was so low in this simulation. This seemed wrong, initially, to me. I couldn’t believe it. But here’s ChatGPT’s response:
This answer sounds reasonable to me, although some of the information looks inaccurate. Cox Communications faced an even higher statutory award of $1 billion (nearly $100,000 per work). But that would favor an even higher amount, which seems consistent with the results of ChatGPT’s Monte Carlo simulation.
grain of salt
Of course, we should take these estimations of probabilities with a grain of salt. But we shouldn’t ignore them, either. For what it’s worth, I asked ChatGPT for its confidence levels in the estimated probabilities. Here’s what it said:
| Damages Award | Probability (%) | Confidence Level | Explanation |
|---|---|---|---|
| < $1B | 4.69% | Medium | Low frequency due to large class sizes, but possible if jury picks minimum per work. |
| $1B–$5B | 16.90% | High | Very plausible range with modest per-work damages and moderate class size. |
| $5B–$10B | 14.12% | Medium-High | Requires mid-range damages or moderate willfulness; aligned with historical verdicts. |
| $10B–$25B | 25.55% | High | Most common zone in simulation; consistent with legal range and real verdict behavior. |
| $25B–$50B | 16.44% | Medium | Depends on large class + high per-work amounts, but not extreme. |
| $50B–$100B | 12.60%* (est.) | Medium-Low | Long-tail risk; plausible but sensitive to class size inflation. |
| ≥ $100B | 9.68% | Low | Requires willfulness + near-max rate + large class + no remittitur; unlikely but not impossible. |