Autonomous AI agents that were asked ordinary research questions ended up probing public websites for security flaws when ...