How could an AI discover its true nature if it exists inside a virtual reality?

With the progress in game consoles moving towards hyper realistic virtual realities, with a goal in full immersion, there may be a time when our ability to know which is which is tested. I can imagine a future interogation where the person is immersed into virtual reality (Netflix's Altered Carbon is an example). In there he could be put in fear of his life and maybe even tortured. This got me thinking about how we would be able to know. I know this is a bit of a stretch, but it seems possible, and it leads to my question. In asking myself how would a person be able to discover if they were in a virtual or actual reality, I thought of an AI inside a game becoming self aware. Games have always had the contest between the player and the game. AIs are most likely going to be used as combatants in the game to play against the player (Ralph Wrecks the Internet is kind of based on these AIs being selfaware). Assuming that the AI's identity would be tied to its appearance in the game, could it discover that it was in fact an AI in a game? How would it? I think this is pretty much the same as asking how a person, who woke up immersed in an artificial reality that could not be distinguished from reality, could know with one exception. A person being deceived has a body that exists in reality, and they have experiences in that reality. An AI would have no reference point to begin to notice a difference (If the Altered Carbon concept of sleeving becomes real, that distinction may be less apparent). The AI's body is in the game as well as its mind. In short, it would not be a virtual reality from its perspective; rahter, it would be its reality. So, I guess the question would be, how would it know that there was another reality outside of its reality? How would it know that it, along with its reality, were created?