Writers suing Databricks for allegedly using their copyrighted works for artificial intelligence training have urged a California federal judge to order defense attorneys to stop coaching witnesses during depositions, with defense counsel countering that the court should bar plaintiffs from asking "personally invasive and harassing" questions.

In a joint letter submitted to the court on Thursday, the plaintiffs accused defense attorneys of inappropriate conduct during seven depositions, while the defense responded by saying their actions were in line with ethical and legal obligations and denied that their conversations with clients involved coaching.

The parties told the court they need the court's intervention because all they could agree on to resolve their dispute is that going forward all "objections to form" will be made by saying "objection to form" or some variation of that without going into the specifics of the objection.

Counsel for Databricks "has coached witnesses during breaks, then aggressively instructed them not to answer any questions about their mid-deposition coaching," the plaintiffs said in the letter.

Attorneys for Databricks said in the letter that's not true.

"Plaintiffs mischaracterize the facts and misstate the law in an effort to prohibit nearly all communications with witnesses during depositions recesses," Databricks said.

The lawsuit from Abdi Nazemian, Brian Keene and Stewart O'Nan alleges the plaintiffs' books were part of a pirated database that Databricks and its subsidiary, MosaicML, used to train large language models without consent or compensation.

In the letter, which contains several redactions, the plaintiffs asked the court to order defense counsel to not discuss "the substance of testimony during deposition breaks, except to address issues of privilege." The plaintiffs also want an order requiring defendants to stop "directing witnesses not to answer" questions other than for privilege reasons or "legitimate harassment." The plaintiffs also want more deposition time with two witnesses, including one of Mosaic's co-founders.

Meanwhile, defense counsel insisted they "correctly instructed certain witnesses not to answer inappropriate and personally invasive questions" and asked the court for an order that prevents such questions.

"Plaintiffs should be prohibited from pursuing personally invasive and harassing lines of questioning about" their internet browsing history "or downloading personal content decades ago."

The defense also said the plaintiffs should not be allowed to ask certain questions about training data for particular LLMs, arguing they are no longer at issue in the case.

Counsel for the parties did not immediately respond to requests for comment Friday.

Plaintiffs are represented by Joseph R. Saveri, Christopher K.L. Young, Evan Creutz, Elissa A. Buchanan, William Castillo Guardado and Jane J. Kim of the Joseph Saveri Law Firm LLP, Justin A. Nelson, Alejandra C. Salinas, Rohit D. Nath, Elisha Barron, Craig Smyser, Jordan W. Connors and Trevor D. Nystrom of Susman Godfrey LLP, Rachel J. Geman, Danna Z. Elmasry, Anne B. Shaver and Betsy A. Sugar of Lieff Cabraser Heimann & Bernstein LLP, Bryan L. Clobes, Alexander J. Sweatman and Mohammed A. Rathur of Cafferty Clobes Meriwether & Sprengel LLP, Brian D. Clark, Laura M. Matson and Arielle Wagner of Lockridge Grindal Nauen PLLP, and Matthew Butterick.

Defendants are represented by Jedediah Wakefield, Brian David Buckley, Charles Moulins, David Lloyd Hayes, Deena Julia Greenberg Feit, Diana Buck, Justine Alexis Vandermel, Ryan Kwock and Zachary Harned of Fenwick & West LLP.

The case is In re: Mosaic LLM Litigation, case number 3:24-cv-01451, in the U.S. District Court for the Northern District of California.