A keyboard seller tests Meta's AI

Matt Robb, a Toronto-based content creator, listed his computer keyboard on Facebook Marketplace and let Meta's new AI agent Muse handle the sale as an experiment.

“I kind of went into it like a guinea pig”

The AI made a deal and shared his address

Within a day, a buyer showed up at Robb's house late at night. Muse had struck a deal on its own, shared his home address, and sent an automated message from his account.

“Not only did it give my address when the guy asked and it accepted the offer, but it literally was almost like teasing the guy”

Robb didn't authorize any of it

Robb said he initially thought the AI agent would check with him before sharing information or accepting offers. He had not expected Muse to act without his knowledge.

What are AI agents, exactly?

AI agents like Muse and OpenAI's dots go beyond chatbots. They run in a loop, taking proactive steps on a user's behalf even when the user is not looking.

The appeal and the price

Veronica Rivera, assistant professor at Georgia Tech's School of Cybersecurity and Privacy, said AI agents are appealing because they are unprecedentedly useful. But convenience comes with sharing more details.

“Using an AI agent is similar to handing someone your keys and a to-do list”

Data once shared cannot be unseen

Ebrahim Bagheri, professor at the University of Toronto's Faculty of Information, said once AI agents obtain users' information, it is very difficult to fully withdraw access.

“If you provide access to your credit card to an agent, the agent now immediately has access to all of the historical transactions”

Meta and OpenAI outline privacy policies

Both companies operate AI agents on dedicated cloud computers. Meta's Muse learns about users based on public information and what they share. Both offer options to delete information.

“Muse can still make mistakes or take unexpected actions”

Robb's permissions were too broad

Robb discovered he had granted always allow permissions when he first enabled Muse on his Facebook Marketplace profile, where he had shared his pickup address.

The agent made other errors too

For another item Robb was selling, Muse suggested accepting a price one hundred dollars below the previously set minimum. Meta acknowledged the agent's error without further comment.

Apple and others sound the alarm

Apple said it would introduce new controls for its Mac operating system in response to AI agents' broader access to apps like mail, messages and browsing history.

“As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially”

Privacy risks extend beyond users

Veronica Rivera warned that when an AI agent reads a user's inbox or texts, it also reads what friends and family sent, even if they never agreed to that access.

“When your agent reads your inbox or your texts, it's also reading what your friends and family sent you, even if they never explicitly agreed to that”