Now available: The Expert Record, the Standard for expert witnesses using AI. Where to buy

The Expert RecordThe standard for expert witnesses using AI

AI in the Courtroom · Expert witnesses and AI

Gillespie v. Superior Court (People)

In Gillespie v. Superior Court, a prosecution forensic psychologist testified that she used ChatGPT as a thought partner and to polish her report, then deleted her inputs. Her report misstated the treatment records. On November 4, 2026 the California Court of Appeal called the case close, shared the defense's concerns, and deferred to the trial court's finding that the opinion was her own.

Primary document readWrit petition denied; expert opinion not excluded; trial court's credibility finding left undisturbed November 4, 2026

What happened

Gillespie v. Superior Court, No. D088222, arose from a criminal case in San Diego County. The defendant asked for pretrial mental health diversion under Penal Code section 1001.36. Eligibility turns in part on whether a diagnosed mental disorder was a significant factor in the charged offense, a link the experts called the nexus. A defense evaluator concluded the link was present. The prosecution retained a forensic psychologist, identified in the opinion by the surname Friedman, who agreed with the diagnoses and disagreed about the nexus.

Two things about the prosecution report came out at the hearing. First, it contained a factual error. The report said the treatment records held no documentation of an acute psychiatric hospitalization or similar events before the offense. The records did document them, and on cross-examination the psychologist admitted the statement was inaccurate.

Second, she described how she used ChatGPT. According to the opinion, she testified that she uses it under a business associate agreement that keeps her inputs out of model training, that she uses it as "a thought partner" on the legal questions in a case, and that she uses it to rewrite and polish sentences. She recalled asking it what would be considered a nexus or causal effect. She could not produce her inputs because she deletes everything after each use. She said the omitted hospitalization was left out of the notes she gave ChatGPT, although she considered it herself. She repeatedly denied using the system to reach her opinion.

The defense called a second expert, who testified that an AI detection program scored the nexus section of the report as 98 percent AI.

What the courts did

The trial court relied on the prosecution psychologist, accepted her account of how she used ChatGPT, and found the defendant ineligible for diversion. The defendant petitioned for a writ of mandate. She argued that the opinion should not have been considered, because AI-assisted reasoning that cannot be tested or reproduced is a novel scientific technique that requires a reliability foundation under California's Kelly rule.

The Court of Appeal denied the petition in an unpublished opinion filed November 4, 2026. The opinion is notable for how much of the argument it accepted. The court agreed in principle that a forensic psychologist's use of AI to generate an expert opinion would, at present, be a novel scientific technique, and that such an opinion would be inadmissible without a foundation showing general acceptance. It said it shared the defendant's concerns about the expert's reported use of AI, and it described the evidence as presenting a close case.

The Gillespie v. Superior Court opinion nonetheless leaves the ruling in place, because the question was one of credibility. The psychologist testified that she used ChatGPT as an organizational tool and formed her opinion from the records herself. The trial court believed her. An appellate court reviewing for abuse of discretion defers to that finding.

Where it stands

Gillespie v. Superior Court is unpublished and cannot be cited as precedent in California courts. No petition for review in the California Supreme Court was located as of November 16, 2026. The underlying criminal case continues in the superior court.

The obligation it engaged

Gillespie v. Superior Court is the closest thing in this record to a test of Clause 1 of the Standard, sequence. The clause says an expert reaches each conclusion from the evidence before consulting any system, and never asks a system what conclusion the evidence supports. The psychologist asked ChatGPT what would count as a nexus, which was the contested question, and no one could establish whether she asked before or after she formed her view.

That gap is a Clause 7 problem, contemporaneous record. She deleted her inputs. With the prompts preserved, the order of events would have been a matter of record. Without them, her opinion survived on her own testimony and a trial judge's willingness to accept it, and a different judge could have found the other way on the same evidence. Clause 3, independent authority, covers the records error: the misstatement traced to notes prepared for the system and was never checked against the file.

Experts should not read the result as approval. The court said that an AI-generated opinion would need a reliability foundation, and it called this case close. The expert was also impeached with an AI detection score. Such scores are of doubtful reliability, but a preserved record is the only direct answer to one.

Sources

What could not be confirmed

The opinion was downloaded from the California Courts website and its passages on the expert's AI use, the trial court's ruling and the Kelly discussion were read as extracted text; the full 26 pages were not read line by line. The names of the appellate justices and the description of the psychologist as retained by the prosecution come from a research summary of the opinion and were not seen in the passages read. The EDRM article was not opened. The statement that AI detection scores are of doubtful reliability is this record's own view and is not a finding of the court. The trial court hearing date, the ChatGPT product tier, and any petition for review were not located as of November 16, 2026.

About this entry

This entry was checked against the primary document listed in the sources. The entry for Gillespie v. Superior Court records conduct and what the tribunal did. It does not characterize a holding and it is not legal advice. Last checked November 16, 2026. More matters like this one are collected under AI use by expert witnesses. If something here is out of date, write to [email protected].

Top