The $1.5 billion settlement by Anthropic over the use of the works of authors to train AI is back on the table. In early September, the record setting agreement was announced by Anthropic and lawyers representing authors but that agreement was rejected by the judge for the case due to multiple concerns.

Now, U.S. District Judge William Alsup has called the class-action settlement “fair” during a hearing that took place on Thursday.

The next step is to notify the affected authors and give them a chance to file claims.

Anthropic is accused of using millions of pirated of books to teach its AI assistant Claude to respond to prompts.

The case has been interesting as Anthropic had been cleared from some of its wrongdoing with the judge ruling that training their model on purchased work was “exceedingly transformative.” Where Anthropic ran into issues was its use of pirated material in that training. About 7 million pirated books were saved to a “central library.”

As reported earlier, the settlement would cover anything done before a certain date but any new infractions after could lead to further lawsuits. Anthropic would also destroy the datasets used in its models. The settlement would amount to about $3,000 per class work.

A trial was set to begin in December over the piracy. Potential damages would have been in the hundreds of billions of dollars.

It was a record setting settlement by Anthropic to authors and it has been rejected by Judge William Alsup with much more to be done. The federal judge has concerns with a $1.5 billion settlement deal and it’ll be forced “down the throat of authors.”

The judge reportedly felt misled by the deal and that it was “nowhere close to complete.” The judge said he was disappointed that counsel have left important questions to be answered in the future.” Questions include some key ones like the list of works involved in the case, the list of authors, the process of notifying members of the class and the claim form members can use to get their settlement payment.

Anthropic was sued for using the works of authors to train its AI model. In the case, U.S. District Judge William Aslup ruled that Anthropic’s use of the books in training models was “exceedingly transformative,” a factor to determine if the use was under legal “fair use.” It was a major decision, the first concerning the issue. But, there was still a question of Anthropic pirating books from the internet to train its models, which would still need to settled in a trial.

In short, they were covered with what they purchased but there was all of the other stuff they pirated that was left to be decided if that was legal or not.

Anthropic in the settlement would have paid $1.5 billion and destroyed datasets used in its models.

About 500,000 authors were involved and they’d have received about $3,000 per work.

The lawyers need to give the class members “very good notice” about the settlement as well as design as claim form where those authors can opt in or out. Anthropic also can’t be sued for the same issue in the future as part of all of this.

The lawyers have until September 15 to submit a final list of works involved in the lawsuit. A class members list and claims form need to be examined and approved by the court by October 10.

After all of that, the court might grant preliminary approval.

(via Bloomberg Law)

As we bring the news of Warner Bros. Discovery’s lawsuit against Midjourney, another case involving AI and theft is being settled. Anthropic has agreed to pay at least $1.5 billion into a class action fund as part of a settlement. The litigation was brought by a group of book authors.

The settlement releases Anthropic for it conduct up to August 25. That means new claims could be filed for any conduct after that date. Anthropic will also destroy the datasets used in its models.

The settlement amount is about $3,000 per class work.

A hearing in the case is scheduled for September 8. A “settlement in principle” was previously announced by Anthropic and the creators.

In the case, U.S. District Judge William Aslup ruled that Anthropic’s use of the books in training models was “exceedingly transformative,” a factor to determine if the use was under legal “fair use.” It was a major decision, the first concerning the issue. But, there was still a question of Anthropic pirating books from the internet to train its models, which would still need to settled in a trial.

Even though they’ll be paying $1.5 billion, it is a win for Antropic and possibly other AI companies allowing them to purchase a copy of the work, and then train their model off of it.

While $3,000 might not seem like a lot, it’s more than the $750 statutory damages amount a jury could award and more than the $200 amount if Anthropic were to prevail on its defense of innocent infringement.