Forum: Author Hangout

Given the size of the fine and their seeming acceptance of it, that does seem a little strange.

That's probably OpenAI's fault. Back in 2023, when the class action lawsuits started, OpenAI freely admitted to the UK government that building AI without scraping copyrighted works would be impossible, and therefore argued for "broad text and data mining exceptions".

So, essentially, the biggest AI company at the time told the world that every good model HAS to have been trained on stolen copyrighted materials. Their last defense was to argue Fair Use, but failed spectacularly with this.

In short... they've been trying to justify their actions for years, and now simply ran out of arguments.

*EDIT*
I also just realized that, if you break the $1.5B down, it's merely three grand per stolen book. That's... relatively little. With the precedent now set, if the authors were to pursue this individually, it could get even more pricey for Anthropic. Because, naturally, they didn't stop scraping materials since the lawsuit started.

so I wonder if there is going to be 'debt restructuring' that enables the fine to be written off against tax.

US tax law specifically blocks deducting legal fines.

It is unusual for large companies (especially ones with in-house legal teams) to accept initial court findings, especially fiscal rulings.

This is not a court finding.

While class action lawsuit settlements have to be approved by the court, the settlement is still negotiated between the named plaintiffs and the defendants without the involvement of the court.

In other words, Anthropic had agreed to the settlement before it was put in front of the judge for court approval.

The court drew a crucial distinction, noting that while the technical training of AI models might qualify as fair use, the unauthorized acquisition and storage of pirated materials is a clear copyright violation.

Seems to me that it was a court agreement, not a pretrial agreement.

Settlements can occur at almost any point in the process (including during appeals made after a judgment has been rendered), so long as the Court approves.

AFIK, even for a settlement made during the appeals process,
court approval is only required for class action lawsuits.

Seems to me that it was a court agreement, not a pretrial agreement.

You seem to have missed this earlier in the OP:

In the largest known copyright recovery in U.S. history, a federal judge finalized a historic $1.5 billion class-action settlement between artificial intelligence company Anthropic and a coalition of authors and publishers.

A settlement, not a court judgement post trial. A settlement by definition is negotiated between the parties outside of court.

Also while settlements can happen before the trial starts, they can also happen mid trial or even while the jury is deliberating.

It's actually pretty hard to imagine a scenario in which this fine results in investors losing money on their anthropic shares? Either the valuation holds, in which case the fine is a fraction of a percent of current value, let alone value at some future IPO date, or the valuation doesn't hold and the fact that anthropic had to pay a 1.5b fine will pale in comparison to the fact that anthropic lost hundreds of billions of dollars in valuation.

It's actually pretty hard to imagine a scenario in which this fine results in investors losing money on their

It's a settlement, not a fine.

It's a settlement, not a fine.

Ah, you're correct. I used the term "fine" because that's how Pixy I described it, but of course they are wrong and incorrect.