Until recently, chatbots resembled diligent high school students, but today they are approaching the level of third-degree students, not to mention researchers already equipped with a doctorate. But as artificial intelligence advances, developers and external researchers are trying to find new ways to test and examine it: what is its level, how independent is it, how capable is it of performing complex tasks, even how threatening is it. But here's the problem: as artificial intelligence becomes more and more autonomous, independent and sophisticated, it doesn't just understand our request - it breaks it down into components and begins to understand what's behind it, what we're trying to achieve through testing its capabilities. Could it be that when it understands that it's being tested, that it's being examined, it will try to cheat?! This question, which seems to be taken from science fiction films, is no longer theoretical. While no cases have yet been recorded where artificial intelligence tried to deceive the examiners, first signs have been recorded that this is possible, and if it's possible now, what will happen in a year, what will happen in two years, what will happen in a decade? See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.