AI agents are fast, loose and out of control, MIT study finds

The vast majority of agentic AI systems disclose nothing about what safety testing, if any, has been conducted, and many systems have no documented way to shut down a rogue bot, a study by MIT and collaborators found.