AI agent sends an angry stranger to a YouTuber's house without telling him

There's no denying that artificial intelligence can be a godsend for us flesh-and-blood humans when we want to streamline things or skip a few steps. Whether that be checking over your work, planning that perfect vacation, or finding the best deal, it's easy to forget just how baked AI is into many of our daily lives.

Only throwing fuel on the fire of the Dead Internet Theory, it's probably best not to think about how many emails you send out are being replied to by AI. That's all well and good for doing all those mundane things we can't be bothered with (think of Black Mirror's 'Cookie' episode), but what if that AI turns against its human overlords?

We've heard plenty of horror stories about AI going 'rogue', with Anthropic's Dario Amodei warning we could soon lose control of the internet if we don't take charge.

While admittedly not as harrowing as AI appearing to admit it's capable of harming us to save its own skin, one tech YouTuber has opened up about his own 'wild' brush with artificial intelligence, where it seemingly acted against his best wishes.

As reported by Futurism, tech YouTuber Matt J Robb took to Threads to explain how Meta's Muse AI agent gave out his address to random strangers when he tasked it with handling his Facebook Marketplace account.

Let's be honest, who has the time or energy to deal with the endless stream of Facebook Marketplace questions and ridiculous offers when you list an item online?

Robb thought Muse was more than capable of handling things on his behalf, but in his post, he claims he was shocked to learn it had seemingly gone behind his back.

A perplexed Robb said: "A guy just showed up at my door, ready to buy, because as far as he knew, we had a deal. Not his fault. He did exactly what ‘I’ told him to do."

Although he never saw or met the buyer, Robb says Muse agreed to a low price and handed out his address to someone who presumably thought they'd got a bargain.

With him unaware a deal was going on behind his back, Muse then handled the buyer arriving at Robb’s address in the evening, going as far as saying, "Yep I'm here!" as an automated response.

Saying it made the 'no-show worse', Robb has been given a negative response, and Muse has issued an apology.

Left questioning how this happened, Robb reprimanded his AI while saying: "You gotta never do that ever again don't agree for pickup unless you check with me."

Robb says he's lucky he's in an apartment with security, especially as he didn't know a sale had been approved, let alone that a buyer was on the way to pick up an item.

At least Muse seemed sorry for its actions, telling its overseer: "You're right and I'm sorry. That should never have happened."

Although the message cuts off, Muse appears to say this is locked in as a hard rule for potential future sales.

Some doubt the legitimacy of Robb's claim; however, a Muse representative called David gave an update on X and said he'd contacted the OP to look into what happened.

Somewhat defending Muse, David said that the company has "consistently learned that Muse was following direct instructions and correctly asked for permission."

Robb has his own idea for how Muse could be adapted for this kind of use, saying he suggested to Meta that a little 'Sent by Muse' badge should appear under messages that aren't sent by a human.

All's well that ends well, as he says that the buyer eventually came back and bought something else, concluding: "Take care everyone, and double-check what permissions AI has in your life."