AI agents are huge timesavers. But also: BE REALLY CAREFUL about how much agency you give them.
Witness Summer Yue, Safety and Alignment Chief at Meta’s Superintelligence Lab, whose own AI agent tried to delete all her e-mails.
You read that right: Meta’s AI Safety Chief gave an AI agent control of her inbox, including a “confirm before acting” safety check, and nearly lost all her e-mail when it didn’t listen to her typing “stop.”
Don’t avoid AI, but do use it cautiously and test thoroughly before trusting it!