Meta’s Muse AI agent faces privacy concerns over its push for personal data, even as Mark Zuckerberg defends the company’s approach to AI safety. Muse, described as a mainstream AI agent programmed to automate personal tasks, has been a hit for the social media giant, working more like a companion than a chatbot with features such as emojis. While testing the agent, WIRED’s Reece Rogers reported continual reminders about connecting more information, with Muse suggesting users add more personal data for it to actively read their financials and other information. The report said the agent’s communication felt like a lure to upload more data, sometimes floating ideas such as keeping tabs on the inbox or taking photos of a meal to estimate food calories. Muse users actively opt to have their data interactions used for AI model training; Meta says data is sanitized to remove identifying information before training AI, though collection might include context from connected sources like an email inbox or a bank account. Users can switch this off in the Muse app’s settings under Data Controls by disabling the toggle that says “Help improve our AI models.” While data might not be shared directly with advertisers, agents browsing on users’ behalf can still see what users are viewing via web-tracking algorithms, potentially leading to influenced ads and marketing campaigns on Facebook Marketplace and other places. Zuckerberg said AI labs have a strong incentive to prevent harm due to “significant” liability, pointing to Meta’s decision to delay Muse’s release earlier this