A climate lawsuit over an aging petroleum storage terminal in Connecticut is emerging as a test case for how courts may treat the use of AI in creating expert testimony.
A federal judge on Wednesday agreed to stay a May court order requiring the Boston-based Conservation Law Foundation to give Shell the details of a search one of its experts used to create a report. CLF had asked for the stay, arguing that the order from a magistrate judge would compel it to disclose material that does not exist — namely, the artificial intelligence prompts that the expert witness used to narrow the scope of her research.
Shell has until June 10 to offer a response. After that, Judge Vernon Oliver of the U.S. District Court for the District of Connecticut will decide whether to keep the stay in place as he reviews the magistrate judge’s order — and then whether to uphold or vacate that order.
The dustup comes as lawyers and their clients increasingly use AI but with little guidance on how courts will consider which AI data may be subject to discovery, said Melissa Weberman, who leads the law firm Arnold & Porter’s eDiscovery & Data Analytics group.
“There’s been sort of this question mark about AI data,” she said. “Some people use it almost like Google and we have never thought that Google entries would be discoverable.”
CLF filed its lawsuit against Shell in 2021, accusing the oil giant of failing to ready petroleum storage terminals in New Haven, Connecticut, for climate-change-induced flooding and severe storms. Shell has called the foundation’s claims “unfounded” and said they should be addressed through the regulatory process, not the courts.
The case has dragged on for years, with a judge in 2023 denying the company’s bid to dismiss the lawsuit and urging the parties to reach a negotiated settlement.
The latest scuffle came after Shell sought material — including AI “prompts and queries” — used by one of the group’s experts, Naomi Oreskes, a Harvard professor who has studied climate disinformation.
Shell noted in a December brief that “we are at a pivotal point in the use of generative artificial intelligence (‘AI’) in litigation.”
The company argued that if Oreskes used AI to narrow the scope of the documents she planned to review, then the prompts she used to set those boundaries should be part of her workflow and made available to the company. It contends that the refusal to produce the information raises “serious concerns” for the company’s ability to cross-examine Oreskes at trial, arguing that it is unable to determine how the data affected her opinions or whether AI hallucinated and provided misleading or inaccurate information.
After oral arguments in the case, Magistrate Judge Thomas Farrish ruled in May that Shell could see the AI prompts because an expert witness’s “methodology is fair ground for discovery.”
Under the facts of the case, Farrish wrote, “the process by which Dr. Oreskes culled down the defendants’ document production into a subset to be worked with is an aspect of that methodology.”
CLF said it welcomed Oliver’s stay of that order.
“Shell’s focus on search terms should not divert attention from the dangers presented by climate change at its unprepared coastal facilities,” said Christopher Kilian, the group’s president for strategic litigation. The foundation, he added, looks forward “to presenting Dr. Oreskes testimony and exposing what Shell knew and the actions they failed to take at their New Haven facility leaving families and businesses in peril.”
Shell declined to comment.
Attorneys who handle cyber and data cases said the ruling is notable because it provides an initial look at how judges may view the use of AI in expert testimony to the courts.
Farrish’s decision is the “very first to rule on the discoverability of AI data in the context of expert discovery,” Weberman said.
With the magistrate’s ruling and in other cases, she said, “the courts are signaling to us that they see AI very differently, that they see this as a data source, like any other. So, whether it’s a text message or an email or a video file, this is just another data source.”
If the order holds, Weberman said law firms may need to instruct lawyers and clients that the directions given to shape AI searches may have to be turned over to the other side, unless the work falls into the traditional categories that are exempt, such as attorney-client privilege and work product protection.
The law firm Alston & Bird echoed Weberman, calling Farrish’s ruling “one of the clearest messages yet on AI in litigation: if an expert used AI to do the work, the prompts may be discoverable.”
The ruling treats AI prompts as “part of the expert’s methodology rather than protected drafting material,” the firm said on its Privacy, Cyber & Data Strategy Blog.
The firm noted that Oreskes had disclosed that she used AI as she reviewed Shell’s production of documents in the case.
But CLF told the court AI prompts were only used to “cull a large document universe” and was “never considered by the witness in forming her opinions.”
It also said that it has “confirmed repeatedly and in good faith” that no data logs keeping track of the prompts were preserved.
Farrish said in his order that if CLF determines “after a diligent search” that no materials exist, “sanctions” may become available if that representation is later revealed to be untrue.
