Meta's Muse AI assistant handles chores well but reveals unsettling depth of personal data

Meta has rolled out Muse, its latest AI agent designed to offload mundane digital chores like email cleanup, shopping, and trip planning. The assistant performed surprisingly well during testing, yet the experience highlighted a troubling aspect: the sheer volume of intimate personal data Muse can tap into directly from Meta's own platforms—information that even users don't readily see.

How Muse performs on basic tasks#

Connecting Muse to Gmail allowed it to successfully filter and delete thousands of promotional emails, demonstrating straightforward capability at cleanup work. When linked to Amazon, the AI intelligently surfaced relevant workout clothing matching specific criteria (size, style, and color), then thoughtfully flagged unrelated items already in the cart before completing the purchase. The tool also generates AI-powered podcasts, images, videos, and interactive documents called "artifacts"—capabilities similar to other agentic AI tools now entering the market.

The assistant did refuse certain image-generation requests (a cartoon mouse with red pants, or a character resembling Mario), but generated product mockups featuring Apple devices without hesitation, complete with hallucinated app names. These performance patterns suggest Muse operates within guardrails, though inconsistently applied.

The privacy wake-up call#

Things turned unsettling when examining what Muse actually knows. After being asked to share information about the user's interests, Muse provided a surprisingly detailed list: anime, CrossFit, Labrador retrievers, Florida wildlife, and '90s/2000s nostalgia. When pressed on the source, Muse revealed it pulls this data directly from Instagram's API—a layer of detail that Instagram's own interface never exposes to users.

"Instagram doesn't show you that exact list anywhere in the app," Muse explained. "I read it through the account's API data, which surfaces more than the UI does." The tool pointed out that users can only see vague ad-targeting categories within Instagram settings, not the granular profiles Muse accesses. The Feed feature added another layer of concern: when asked to generate a personalized news digest, Muse pulled information about the user's current location from an Amazon shipping address, then surfaced hyper-local news accordingly.

What comes next#

Meta maintains that Muse only exchanges data with third-party services "that's needed for them to work for you" and does not share user information with advertisers. Still, the company's track record on privacy—marked by past settlements involving data breaches, unauthorized access, and misuse of sensitive information—makes such assurances difficult to accept at face value.