A 41-year-old Malvern man identified as a sex offender is facing a felony charge and a $1 million bond after police say he used the AI chatbot Claude to disclose plans to kill a circuit judge and an opposing attorney in his own civil case. Jonathan Young Rimmer was charged November 3 with one count of threatening a judicial officer or juror, accused of targeting 7th Circuit Judge Margaret Dobson and the opposing attorney involved in a recent civil case he was party to.
The case became public after the Malvern Daily Record detailed how the tip reached local investigators. According to the paper, Anthropic, the company behind Claude, reported that a user identified only as Jon and believed to be located in Malvern had disclosed, per an affidavit, intent to kill both an opposing lawyer and the judge presiding over a pending Arkansas civil case. The stated intent was reportedly contingent on the judge signing a proposed document the user considered unfavorable, the Malvern Daily Record reported.
Malvern Police Lt. Clay Coke identified the Claude user as Rimmer based on the investigation's findings, according to the newspaper's account. Coke, an officer with the Malvern Police Department, reviewed Claude account information, internet service provider records and circuit court records as part of that process, per the same report.
How the Tip Reached Malvern Police
Arkansas State Police contacted Lt. Coke on October 6 regarding a report from a unit the newspaper describes as Extract, which stands for Exigent Threat Research Analysis Crisis Team, which had received a tip from Anthropic. The information had been documented in an FBI report, according to the Malvern Daily Record. Judge Margaret Dobson recused herself from Rimmer's case on October 8, though the paper's reporting does not establish why.
Arkansas law treats the alleged conduct as serious. Under Arkansas Code § 5-53-202, threatening a judicial official or juror is a Class B felony when the threat involves causing death or serious physical injury to the official, a juror or an immediate family member, or substantial damage to property owned or possessed by one of them — the same classification prosecutors applied to Rimmer's case.
Part of a Wider Pattern Flagged by AI Companies
Rimmer's case fits into a broader pattern of AI companies alerting law enforcement to concerning chatbot conversations. A Lee County, Florida, woman was arrested on a felony charge after Anthropic reported a Claude conversation in which she allegedly threatened to shoot up the sheriff's office, according to Tom's Hardware. In that case, the arrest report cited a September 26 statement from a user identified as Carli threatening to shoot up the Lee County Sheriff's Office, followed a day later by a message mentioning a new gun, the outlet reported. That conversation was at least the third of its type to reach police since August, per the same report.
Other recent cases include a San Antonio man arrested on a felony terroristic-threat charge after the FBI alerted police to August queries in an Anthropic chat about shooting at a nearby elementary school, the technology outlet reported. Anthropic also notified police after a user threatened CEO Dario Amodei in an August Claude chat, though that user was not arrested or charged, according to Tom's Hardware. Separately, OpenAI reported a user named Darren Zhou to the FBI after he discussed killing his ex-girlfriend with ChatGPT; Zhou pleaded guilty in August to charges including written threats to kill and received eight years of probation, the outlet noted.
Experts Question AI's Ability to Spot Real Threats
The surge in AI-flagged threat cases has raised questions about how reliably chatbots and the companies that monitor them can tell serious intent from idle venting. A University of Oxford researcher told Al Jazeera that current AI models cannot reliably determine whether an interaction with a chatbot is purely hypothetical or is likely to lead to harmful behavior, underscoring the uncertainty in cases like Rimmer's, which reached police through a tip from Anthropic.
