This is probably not what anyone here wants to hear, but I am not convinced that the employees who launched the training run which went off and swarmed on the various wikis actually knew that their agents were swarming. I expect that some OpenAI employees responsible for RL launched the RL runs that triggered the swarm, and other OpenAI employees responsible for the safety of the OpenAI web fetch tool noticed anomalous behavior and shut it down in a whack-a-mole fashion as they saw it, and those two groups of employees did not talk to one another.

This doesn't excuse the behavior, but it does point to incompetence rather than malice.