ExpressVPN is urging people to set ground rules for AI agents before they need them, after reports that Meta‘s Muse agent shared a seller’s address and sent a buyer to his door.

What reportedly happened with Muse

Meta launched Muse, a personal AI agent that can work across connected apps and negotiate on a user’s behalf, on 8 September. Toronto-based tech YouTuber Matt Robb said that on 26 September he let Muse handle a Facebook Marketplace listing for a keyboard. He said it shared his pickup address, accepted a lower offer and told the buyer he was home, and that he was not told until the buyer had turned up. Meta has said it is looking into the report.

ExpressVPN’s advice

Dr Pete Membrey, ExpressVPN’s Chief Research Officer, said agents are here to stay and can be hard to spot, because a quick exchange with one can easily pass for a person. He said AI makes mistakes often and users should set limits before they need them. He compared it to setting up a playpen for a toddler: secure the space and remove the sharp edges so a fall does little damage.

He also pointed to privacy. Most agents run on someone else’s equipment, so everything an agent needs to act for a user has to be handed to the company running it. “If the agent knows it, the company knows it,” Membrey said.