Sure, Meta’s AI Muse works, but it sure creeps me out

Details

Meta launched Muse on September 10, 2026, positioning it as the company's first genuine AI-powered productivity assistant after years of entertainment-focused AI features. Muse operates via a cloud-based virtual computer, allowing it to act on your behalf across third-party services like Gmail and Amazon — connecting accounts and executing tasks without step-by-step user guidance. In The Verge's hands-on test, Muse successfully deleted thousands of promotional emails after connecting to Gmail, and navigated an Amazon purchase independently — noticing and flagging existing cart items before checkout. It also generates AI-narrated podcasts (similar to Gemini NotebookLM), images, videos, and interactive artifact documents. Some generation requests were blocked (characters resembling Nintendo-adjacent IP), while Apple product imagery including a foldable iPhone mockup was generated without restriction. The experience was overshadowed by Muse immediately surfacing granular personal interests apparently drawn from Instagram activity — before the reviewer had specified preferences. Meta states it only exchanges data with third-party services 'that's needed for them to work' and does not share user data with advertisers, but given Meta's privacy track record, skepticism is understandable. Muse enters a crowded AI assistant market competing with Gemini, ChatGPT, and Claude. Meta's unique advantage is its massive social graph and behavioral data — which is simultaneously its biggest selling point and its deepest privacy liability.