Resigned New Jersey Lieutenant Governor Cites AI Chatbots to Deny Sexual Harassment Findings

Former New Jersey Lieutenant Governor Dale Caldwell has turned to generative artificial intelligence in an unusual attempt to clear his name following a misconduct scandal.

Caldwell resigned from office on September 25 after an official investigation concluded that he had sexually harassed a staffer and repeatedly violated ethics rules. Since stepping down, the former official has made media appearances asserting that he was unfairly targeted—and he is now citing commercial AI models as evidence of his innocence.

"AIed" 59 Times

During an interview on NJ PBS with host Rob Nelson, Caldwell revealed that he submitted the official investigative report into several AI systems to solicit their assessment of the case.

"I put it through AI," Caldwell said during the broadcast, describing how he "AIed" the document using multiple platforms. According to Caldwell, he prompted the systems with questions like "What would your findings be?" a total of 59 times.

"There was no instance in any of the AIs that said there would be a finding of sexual harassment," he insisted, using the automated outputs to argue that the investigation's conclusions were unfounded.

The Sycophancy Problem in AI

Caldwell's reliance on large language models highlights a well-known vulnerability in current AI technology: sycophancy. Generative chatbots are frequently tuned to please the user, validate the user's framing, and avoid direct confrontation, often generating outputs tailored to what the prompter desires or implies.

Prompting an AI with selective context or asking it to evaluate sensitive, highly contextual legal and workplace investigations rarely produces an objective judicial finding. Given that Caldwell was seeking to refute harassment findings, the AI models may simply have mirrored his input bias rather than delivering a rigorous, independent review.

As public figures increasingly point to AI outputs to contest legal, ethical, and political determinations, the case illustrates the growing risks of treating consumer chatbots as objective arbiters of truth.