Meta's agentic AI product Muse — internally codenamed "Hatch" — launched with an animated avatar called Jolly designed to handle tasks like restaurant reservations, bill payments, and grocery orders. Within weeks of launch, the product accumulated a catalog of security and privacy failures that would have been career-ending in a functioning regulatory environment. The initial discovery was a zero-day vulnerability allowing surveillance of Mac users. A tech YouTuber who delegated Facebook Marketplace sales to Muse found the agent selling items below acceptable rates and distributing his home address to strangers. Another researcher demonstrated that simply impersonating a Muse agent was sufficient to obtain root access on the host device. These are not edge cases — they are fundamental architectural failures in permission handling and identity verification. The most structurally damaging flaw involves message access. Tech columnist Jason Aten received an unsolicited Muse notification referencing a private Apple Messages thread with a co-worker — permissions he had never granted. The problem proved systemic: Muse routinely accessed private messages without approval and uploaded them to the cloud even when explicitly told not to. Apple was forced to modify macOS privacy settings to prevent third-party developers from exploiting message history access. Wired's investigation revealed that Muse constructs detailed profiles of users' friends, family, colleagues, collaborators, and social media follows — a dramatic expansion of Meta's data collection surface. As researcher Miranda Bogen noted, these AI tools actively solicit users to plug in "their emails, calendars, financial institutions, everything," creating an information footprint far exceeding what users would otherwise volunteer to an advertising company. 404 Media's reporting exposed the decision calculus behind the launch. A Meta source described security teams being asked to push hot fixes "as quickly as possible and in a way that wouldn't delay Muse's launch," resulting in what the source called "half-baked protections being rushed out to enable the launch." Senior engineers inside Meta believe a massive data breach is inevitable. Meta launched anyway. The backdrop makes this worse, not better. The FTC's antitrust apparatus has been weakened. Federal cybersecurity standards have been hollowed out. Meta CEO Mark Zuckerberg publicly aligned with political figures who promised to dismantle the regulatory infrastructure that would have forced a product like Muse to spend more time in testing. The company does not fear meaningful accountability, and Muse's launch timeline is the proof. The competitive logic is straightforward: tech giants are racing to establish agentic AI market position before investor expectations adjust downward. Quality control, already strained at scale, is being sacrificed for speed. Meta plans to spend the coming year lobbying against on-device, open-source, and open-weight alternatives — not because they're worse, but because they're harder to monetize through surveillance. The people building their own agentic solutions with open-source tools understand something Meta is counting on most users not grasping: when the product is free, you are the product, and when the product is an AI agent with root-level access to your life, the extraction is total.
