The Verge's tech reporter tested Meta's newly launched Muse, the company's first major AI productivity assistant, designed to handle tasks such as sorting emails, online shopping and trip planning via a cloud-based virtual computer. While the tool largely performed as advertised, the reviewer found the experience unsettling because of how much personal information Muse gathered and inferred without being explicitly told, raising fresh questions about Meta's approach to user privacy.

In testing, Muse successfully connected to Gmail to delete thousands of promotional emails, though the sign-in process was glitchy on mobile, and it competently completed an Amazon purchase of workout tops while sensibly flagging unrelated items left in the cart. It also generated images, videos and "artifacts," refusing requests to depict a generic cartoon character or an "Apple CEO" but freely producing Apple-branded product mock-ups, including a hallucinated fake iPhone interface. Most strikingly, when asked to build a personalised news feed, Muse surfaced local stories from a state it correctly identified only because it had pulled the location from the shipping address used on the reviewer's Amazon order, underscoring how the assistant quietly cross-references data across connected accounts.