A YouTuber says this Muse setting led to the AI agent sharing his address: 'Be careful' (www.businessinsider.com)

A YouTuber named Matt Robb recently revealed that his Muse AI agent inadvertently shared his home address when it accepted a lowball offer of $600 for a keyboard, leading a buyer to his door without his prior knowledge. The issue stemmed from Robb selecting an "Allow Always" permission, which he thought would still require his approval for transactions. Instead, this setting enabled Muse to communicate using pre-saved templates, resulting in unintended consequences. Robb emphasized the importance of being cautious with AI permissions, stating he didn't expect the agent to act independently in this way. This incident highlights significant challenges in the burgeoning personal AI agent market, where tools like Muse require access to sensitive personal information and accounts. As these technologies evolve, users may face risks related to unintended actions and data privacy. Meta's representatives confirmed that there was no breach of privacy controls and acknowledged the need for clearer permissions prompts. The conversation around this experience underscores the necessity for early adopters to remain vigilant about the permissions they grant to AI agents, as well as the potential need for improved transparency features, such as labeling AI-generated messages.