When you’re conducting an internal investigation and using AI as a tool, you’re risking the use of a shifty informant.

Let’s talk about a risk that I’m seeing firsthand in internal investigations: AI hallucination.

I use AI in my own practice, and I got scared very quickly because it cited cases that don’t exist. It cited them confidently, persuasively, like it was reading straight off a court docket. It wasn’t.

Now imagine that inside an internal investigation. You’re using an AI tool to summarise witness interviews, to search a document universe, to draft findings for audit committees.

If that tool fabricates a fact, misattributes a quote, or invents a document that was never produced and nobody catches it, that error doesn’t stay contained.

It becomes your investigative record. It becomes the basis for a disciplinary decision, a self-disclosure to a regulator, maybe even a certification to the government that your findings are accurate.

Here’s the rule I put in place today: every AI-assisted output in an investigation gets independently verified against the underlying source material.

Not spot-checked. Verified.

Because the day you rely on an AI summary you never checked is the day that your investigation’s credibility, and maybe your own, is on the line.

The Ethics and Compliance Q and A show is produced by One Stone Creative.