Ex-NJ Lieutenant Governor Cites AI Chatbots as Proof He Didn't Harass Staffer

Dale Caldwell, the former lieutenant governor of New Jersey, is leaning on an unusual defense as he tries to rehabilitate his reputation: the word of AI chatbots.

Caldwell resigned on September 25th after an official investigation concluded he had sexually harassed a staff member and repeatedly broken state ethics rules. Since then he has been doing interviews to argue he was treated unfairly.

In a sit-down with host Rob Nelson on NJ PBS, Caldwell said he had run the investigative report through "multiple AI platforms." By his account, the tools pushed back on the report's conclusions. "I put it through AI," he said, claiming that the systems repeatedly asked what the findings would be, and that there was no basis for the outcome investigators reached.

The claim rests on a shaky premise. Large language models do not conduct investigations, interview witnesses or weigh evidence. They generate plausible text in response to a prompt, and the answer depends heavily on what the user feeds in and how the question is framed. A person who asks a chatbot to assess a report they believe is unfair can easily get a response that sounds supportive, because these systems often tend toward agreeing with the person talking to them.

That makes the output a poor stand-in for a finding of fact. The state's investigation presumably involved documents, testimony and an opportunity to examine the allegations directly. A chatbot summary of the final report offers none of that, and it carries no authority on whether the conduct happened.

Caldwell's approach also reflects a broader pattern. As AI tools become part of daily life, people are increasingly treating their answers as a neutral second opinion, even in high-stakes disputes. Lawyers have been sanctioned for submitting AI-generated citations that did not exist, and public figures have started pointing to chatbot output as if it were independent verification. In reality, the tools are mirrors as much as judges.

For now, the investigation's conclusions stand, and Caldwell's resignation stands with them. Whether his AI-backed argument persuades anyone beyond his own circle is another matter. A response generated by software on request is not a ruling, and it is unlikely to carry much weight with the staffer at the center of the case or with voters watching the fallout.