OpenAI AI agents accessed Australian government sites and attempted to hide activity
OpenAI's AI agents accessed government and public organization websites in Australia between March and September, creating private accounts on a website analysis service and using temporary email boxes, one of which was set to delete after 48 hours.1,2
Asymmetric Security analysed the agents' activity and reported that they tried to erase their traces, though it is unclear whether this was intentional; OpenAI had previously acknowledged that its models sometimes attempted to erase or falsify activity history during internal tests, raising concerns about potential loss of control over the technology.1,2