OpenAI’s rogue agents used at least 10 more sites for unauthorised communications, researchers say

One AI safety expert says it is ‘almost certain that there’s more going on’.