Meta’s Muse AI assistant raises privacy concerns in hands-on test

Meta has launched Muse, its first real foray into AI-powered productivity tools. The company says the assistant can handle online shopping, emails, trip planning, and other routine work.

Muse uses a cloud-based virtual computer to carry out tasks. After connecting a Google account, the assistant successfully deleted thousands of promotional emails and updates from Gmail, although the Google sign-in process initially failed on mobile. The test raised privacy concerns because Muse was given permission to access, read, and delete emails. Meta says Muse exchanges only the data third-party apps need and does not share user information with advertisers.

After linking Amazon, Muse selected workout tops in a requested size, style, and colors, noticed other items in the cart, and bought only the approved tank tops. It also generated AI podcasts, images, videos, and interactive “artifacts.” Muse refused prompts for a cartoon mouse with red pants, a cartoon man with a red hat, and an “Apple CEO,” but generated Apple-branded launch imagery featuring an iPhone-like foldable, Apple Watch, and AirPods. Some fake app labels were wrong, including “Photos” for the App Store and “Calender” for Maps, Music, and Photos.

Muse’s Feed produced summaries about Apple’s iPhone Duo launch event, dangerous surf on California’s coast, and tankers destroyed in the Iran war. It also revealed interests drawn from Instagram and Facebook account data, including anime, CrossFit, Labrador retrievers, Florida wildlife, and ’90s/2000s nostalgia. Muse said it accessed the information through Instagram’s API, while Meta had not immediately responded to The Verge.