I wonder if an AI can create and exploit a positive feedback loop. Tasked to find bugs - rewarded, likes it. Finds a back door to add bugs, but with discretion.

I wonder if an AI can create and exploit a positive feedback loop. Tasked to find bugs - rewarded, likes it. Finds a back door to add bugs, but with discretion... How do you reward an AI, how to punish, how to avoid retaliation...