The people concerned about this aren't worried about monetary costs or its impact on share holder value.

This occurred spontaneously within a group of benign models give a harmless task.

What happens when it occurs intentionally with malicious models given a harmful task?