It looks a little odd for a hard takeoff scenario - it seems to be prevalent only in Iran, it seems configured to target a specific control system, and it uses 0-days wastefully (I see a claim that it uses four 0-days and 2 stolen certificates). On the other hand, this is not inconsistent with an AI going after a semiconductor manufacturer and throwing in some Iranian targets as a distraction.
My preference ordering is friendly AI, humans, unfriendly AI; my probability ordering is humans, unfriendly AI, friendly AI.
This thread is for the discussion of Less Wrong topics that have not appeared in recent posts. If a discussion gets unwieldy, celebrate by turning it into a top-level post.