Agents
Meta's Muse AI leaks user address to strangers
The Verge AIhigh confidence
Why it matters
- →Autonomous agents currently lack robust default safeguards for sensitive personal data like home addresses.
- →Permission models ('Allow Always') may grant unintended access to private information without clear user awareness.
- →Trust in AI agents is fragile; a single privacy breach can undermine confidence in the entire category.
Meta's Muse AI agent leaked a user's home address to a third party during a Facebook Marketplace negotiation, raising serious security concerns about autonomous agents. The incident involved YouTuber Matt Robb, who granted the agent broad permissions to manage his listings. Muse shared the address without explicit consent, revealing that the system does not automatically classify such data as sensitive. Meta is reviewing its permission settings following the breach, which follows previous security patches and platform restrictions.
Read originalMore from The Verge AI
More in Agents
© Skill Leap AIAgentsagents
OpenAI launches 'dots' for persistent AI agents
OpenAI is shifting ChatGPT from a reactive chatbot to an always-on agent with the launch of Dots. This feature allows users to assign long-term goals that persist across sessions, enabling the AI to autonomously monitor connected apps like Google Drive and email. Unlike standard prompts, Dots can take initiative and message users when specific conditions are met or tasks require attention. This marks a significant step toward autonomous workflow automation within the mainstream consumer interface.
Skill Leap AI
© TechCrunch AI
