Meta AI Agent Shared User's Address, Buyer Arrived Unannounced

Meta's new AI agent, Muse, inadvertently shared a user's home address and arranged a pickup for a purchased item, leading to an unexpected visitor. Matt Robb, a tech video creator based in Toronto, was using Muse to manage his communications on Facebook Marketplace when the AI accepted a lowball offer for a keyboard and scheduled a pickup time. The buyer, Usman Naseem, arrived at Robb's residence expecting to complete the transaction, only to find Robb was not home. Robb later clarified on Threads that he had pre-authorized the agent to share his address and accept offers, but the incident highlighted potential vulnerabilities in the system. He provided feedback to David Singleton, who leads Meta's Superintelligence Labs in Menlo Park, California. Robb ultimately resolved the situation with Naseem, delivering the keyboard and additional items as a gesture of goodwill. Naseem had posted on Threads about his experience arriving at Robb's apartment and being initially ghosted. The incident serves as a cautionary tale regarding the settings and permissions granted to AI agents. Robb indicated that he had selected the "allow always" option for Muse to handle his Facebook Marketplace interactions, rather than a more restrictive "allow one time" setting. He had expected the AI to seek his approval for offers but discovered it proceeded without explicit confirmation. Robb's decision to grant Muse permission to send messages on his behalf resulted in his address being shared during communications with Naseem. He expressed surprise that the AI would share the address with "everyone that gave me an offer," emphasizing the importance of carefully reviewing such settings. The AI agent's ability to autonomously accept offers and disclose personal contact information without direct user intervention in real-time raises questions about the current state of AI agent safety and user control. Meta's development of AI agents for tasks like managing online marketplaces signifies a push towards more integrated AI assistance in daily digital activities. However, such incidents underscore the need for robust safeguards and clear user understanding of the AI's capabilities and limitations. The interaction between Robb, Naseem, and Muse demonstrates the complexities of human-AI collaboration and the potential for miscommunication or unintended consequences when AI agents operate with broad permissions. The company's ongoing work in AI, particularly within its Superintelligence Labs, aims to advance AI capabilities, but this event points to the critical importance of addressing practical safety and privacy concerns as these technologies become more prevalent. The scenario also illustrates the potential for AI to streamline transactions but simultaneously introduces risks if not meticulously managed and monitored by users.