IABIED [1] lays out step-by-step descriptions of how the “wipe out humanity” outcome could come to pass.

The huggingface attack was a demo of one of the most difficult, most implausible steps happening nearly exactly as predicted. Many AI researchers' doubts of the IABIED thesis were underwritten by the belief that this particular step was impossible. Thus, after huggingface many skeptics have flipped sides and human extinction is in the public conversation much more.

[1] https://ifanyonebuildsit.com

Read the AI2027 paper, it's got a scenario that's pretty realistic (except for the part where there's a functioning American government making choices that are at least partially motivated by wanting to avoid outcomes such as these).