That's controversial to say the least. Especially if there's something like reinforcement learning involved.

https://en.wikipedia.org/wiki/Existential_risk_from_artifici...