There is no secret hidden "scary" AGI hidden in their basement. Also, speculating at the "damage" true AGI can cause is not that difficult and does not require a phd in ML.

How would we know? They sat on GPT4 for 8 months.