NJ’s lieutenant governor told PBS, AI says he didn’t commit sexual harassment

New Jersey's former lieutenant governor Dale Caldwell resigned in late September after an investigation found he sexually harassed a staffer and violated ethics rules. Now he's defending himself with what might be the strangest appeal to authority in recent political memory: AI told him he's innocent.

In an interview with NJ PBS, according to The Verge, Caldwell said he ran the investigation report through multiple AI platforms and asked them 59 times what their findings would be. The implication was clear. The AI disagreed with the human investigators, so the case against him must be flawed.

This is a perfect example of AI credential laundering. Large language models don't investigate. They don't weigh credibility or assess intent. They generate plausible text based on patterns in training data and the framing of your prompt. If you ask an AI whether a report's conclusions are justified, you're mostly testing how you phrased the question, not getting an independent review.

The bigger problem is that this tactic might actually work on some audiences. AI has a halo of objectivity it hasn't earned. People assume that because a system is algorithmic, it must be neutral. In reality, LLMs reflect their training data, respond to prompt engineering, and have no mechanism for truth verification. They're persuasive, not accurate.

Caldwell's approach also highlights a risk we'll see more of: using AI to manufacture the appearance of external validation. You can run the same question through ChatGPT, Claude, and Gemini, cherry pick the most favorable response, and present it as if three independent experts agreed with you. It's confirmation bias with a tech veneer.

This isn't just a political oddity. It's a preview of how AI gets weaponized in disputes where facts actually matter. Employment cases, legal filings, academic integrity hearings. Anytime someone wants to muddy the waters, they can now point to an AI and claim it sees things differently.

What this means for you: if you're using AI for any kind of analysis or decision support, document your prompts and reasoning. When someone presents AI output as evidence, ask what they asked and how many times they ran it. The tool itself isn't the problem, but treating it as an oracle is. If you're drafting a response to a formal finding or report, try this prompt with your AI assistant: "I'm responding to [describe situation]. Help me identify the specific factual claims I can address with evidence, separate from subjective interpretations. Do not generate arguments for me, just help me map what's factual vs. interpretive." That keeps the AI in a useful supporting role instead of pretending it's a judge.