Meta's Muse: Cute AI Agent Or 'AI Trojan Horse'? | BBC News

0:01 – 10:57

A cute face on a very invasive idea

The show opens with a contrast that does most of the work by itself: while other AI companies are talking about danger and pulling back launches, Meta is charging ahead and putting a smiling face on its newest product. Mark Zuckerberg calls Muse the center of Meta's vision, a "personal super intelligence" for billions of people. That framing matters, because the host immediately flips the question from inspiration to intrusion. Muse is not presented as just another chatbot. It watches what you do, learns how you think, and acts for you, first on the desktop and soon through physical devices like a keychain or pendant. Ed Zitron gives the blunt version of what Muse is in practice. He says it has its own virtual machine with about 8 GB of RAM and can do calendar changes, emails, web browsing, and certain website actions. But his more memorable point is that this is the same old AI promise with more risk attached. He cites a case where someone asked Muse to run their Facebook Marketplace interactions and the system gave a random person the user's home address, leading to "a bloke" showing up at the door. That story lands because it turns an abstract concern, hallucination, meaning the model confidently inventing false information, into a front-door problem. Stephanie Hare explains why Meta gave the agent a face and a name. People naturally humanize tools, she says, and making AI cute stops users from thinking about what it is really doing. Her analogy is sharp: calling them cookies instead of surveillance trackers. The pendant and wearable angle makes that concern more intense. A screen with cameras on the front and back, worn on the body, starts to look less like an assistant and more like a companion that is always collecting. The discussion then widens from design to business model. If Muse can clear your inbox, book your table, buy products, and answer messages, then the real question becomes who is choosing on your behalf. Stephanie calls Meta's strategy a toll booth model. Zuckerberg has openly said Meta would take a small fee from every transaction, booking, and purchase. That means the product is not just helping you move through life more smoothly. It may be positioning itself between you and every action you take, taking a cut each time. The panel keeps returning to one uncomfortable fact: Meta is asking for a remarkable amount of trust. The host cites a Surfshark finding that Muse could in theory collect 31 of Apple's 35 data types, including precise location, financial information, health information, contacts, emails, and browsing history. Ben Hatton pushes back slightly by noting that consumers have been handing data to Apple for years, and he adds an important technical distinction from his Qualcomm conference: in theory, powerful on-device chips could keep more of this processing local instead of sending everything to the cloud. But everyone agrees that is not Meta's actual model here. This system is useful precisely because it can see everything, remember everything, and connect to commerce. That is what makes it feel magical, and what makes it feel dangerous.