Surfshark: Muse Collects 31 of 35 App Store Data Types, Second Only to Meta AI
Surfshark's analysis of Apple App Store privacy labels puts Meta's Muse assistant at 31 of 35 possible data types, far above the 13-type average.
By Icarus4 min read
Surfshark: Muse Collects 31 of 35 App Store Data Types, Second Only to Meta AI
Surfshark’s study of Apple privacy labels puts Meta’s AI agent well above the 13-type average across chatbots. It also finds Muse interactions feed Meta’s AI training by default, with opt-out in the app’s settings.
Meta’s Muse lists collection of 31 out of 35 possible data types in its Apple App Store privacy disclosure, the second-highest total among the 13 AI chatbots analyzed by VPN company Surfshark, behind only Meta’s own Meta AI at 33. The study, published September 28, puts Muse’s disclosed data appetite at more than double the 13-type average across the apps surveyed, and well above ChatGPT’s 17 and Google Gemini’s 24.
Surfshark compared the App Store privacy details of 12 leading AI chatbots, selected from Zapier’s 2026 best-chatbots list, and added Muse after the agent topped the free iPhone charts in its first week, capturing the labels on September 22, 2026. Apple defines 35 data types across 16 categories, and developers self-report which types their app may collect.
How the apps compare
By number of data types disclosed, Meta AI leads at 33 of 35, followed by Muse at 31. Google Gemini discloses 24, ChatGPT 17, DeepSeek 14, and Pi 3.
ChatGPT’s count is up 70 percent from the 10 data types Surfshark recorded a year earlier, with health and fitness, search history, audio, and advertising data among the additions. Gemini added one type, audio, since the previous review.
The categories matter as much as the counts. Muse is one of three apps, alongside Meta AI and Gemini, that disclose collecting what Apple defines as “sensitive information”: racial or ethnic data, sexual orientation, pregnancy or childbirth information, disability, religious or philosophical beliefs, trade union membership, political opinion, and genetic or biometric data. Muse is also one of four apps, with Meta AI, Gemini, and Perplexity, that list precise location; most competitors disclose only approximate location or none.
Two findings are specific to Muse. Meta uses Muse interactions for AI training by default, though users can opt out in the app’s settings, per Meta’s Muse privacy policy. For what to check before you grant an agent access to anything, see our beginner’s guide to AI agent privacy and safety. And like Meta AI and Gemini, Muse reserves the right to use collected audio data for purposes beyond functionality, such as product personalization or analytics, while ChatGPT and DeepSeek say audio is collected for functionality only. Microsoft Copilot was the only app in the group that disclosed user tracking under Apple’s definition, which covers linking app data with third-party data for advertising or data-broker sharing.
Labels are disclosures, not proof
One caveat shapes how to read the ranking. App Store privacy labels are self-reported: developers declare which data types their app may collect, and the categories are broad. The study compares disclosed appetite, not confirmed behavior. A high count means the developer claims the right to collect, not that every type is collected from every user. Surfshark’s own methodology note says the data comes from the labels as published.
The labels are the one standardized, public window into what AI apps say they collect, and the pattern across agents is consistent. Surfshark argues that AI agents are structurally more data-hungry: to act on a user’s behalf, an agent needs access to calendars, email accounts, financial tools, and other connected services that a chatbot answering questions never touches.
Why it matters for Muse users
The study lands in a week of mounting scrutiny of Muse’s data handling. Apple said it would begin flagging AI requests for Mac system data after complaints tied to Muse’s behavior on Mac and iPhone, and reporting this week showed Muse is designed to build persistent profile pages for the people in a user’s life. The privacy labels add the broadest measure yet of what the app says it may take in.
Users have two practical levers from the study. The AI-training opt-out lives in the app’s settings, and the labels are worth a look before connecting an agent to email, calendars, or bank accounts. An assistant that acts with your permissions reads its data disclosure differently than a chatbot that only answers.
