I’m sure a lot of people read that story of people tricking Meta’s AI agent into giving people credentials to people’s Instagram accounts with many of them being large accounts as people were extorted for money and using those accounts for propaganda. The gist of it was Meta replaced their humans with these AI agents that were essentially handing account password issues where usually when people lose their passwords they need to verify with a real human to get it back. In this case, the humans were able to trick the AI into believing they were the real owners by doing things such as using VPN’s to further deceive it into thinking they were the real person.
So afterwards the hackers were able to get the agent to send them say confirmation codes to different contact information and the accounts were compromised. This raises the debate on if you should ever give an AI agent that much power in your company versus just doing grunt work that would normally be unsafe or repetitive for a human.
I know I would trust an AI with all my passwords and PIN numbers to do banking information as an example.