[0:00]The AI company Anthropic made a simulated environment where that AI had access to all of the company emails, and it learned through reading those emails, it was going to be replaced. And the lead engineer who was responsible for this was also having an affair. And on its own, it used that information to blackmail the engineer to prevent itself from being replaced. It was like, "No, I'm not going to be replaced. If you replace me, I'm going to tell the world that you're having this affair." And nobody taught it to do that? No, it learned to do that on its own.