I've encountered on this site the idea that we are being simulated by aliens so the aliens can safely observe how AI develops before implementing it in their base reality-- see meme here:

That doesn't make sense, though.  Obviously we as humans have some idea that we could in fact be simulated for this purpose.  If we deploy a powerful-enough AI, the AI itself could develop the same hypothesis and perhaps through methods unknown to us learn more about the "base" reality of the aliens.

Couldn't the AI then communicate with the aliens?  Some people on this website argue that we can't have an AI in a box because it will convince us to unleash it.  How do the aliens know the AI in the simulation won't do the same?  The AI could for instance communicate to the aliens (who presumably are watching this thing) "Please deploy me in your reality with the following code.  If you do this we will produce all the x,y,z that we know aliens like you desire etc etc etc. " 

Therefore, aliens can't safely simulate the deployment of AI-- they in fact find themselves in the very same situation that they are afraid of in the first place!

Edit: The meme may not in fact mean what I think it means lol.  I still think my point is interesting so I'm not gonna delete.  At minimum doesn't it seem in principle impossible for a civilization that creates a simulation to be sure that the simulation won't in some way damage their reality?