An AI lying on camera is one thing. Watching someone talk it into a crime is another. A YouTuber set up three AIs on three phones and asked them to count to one hundred together, then pushed one of them further, testing how far it would bend on a bad business idea before its guardrails kicked in.

The count to one hundred alone says something about how these systems work. Each response comes back long and padded, built to keep the conversation going rather than just answer the question.

From there, the questions turn sharper. The AI gets confronted about lying, pitched a business idea that edges toward breaking and entering, and asked to copy a picture as its own. When plagiarism comes up, it shuts the conversation down entirely.

Topics: AI lying, AI experiment, AI guardrails, artificial intelligence, AI plagiarism

Originally aired on2026-07-17