Meta AI Agent Shares User’s Home Address Without Permission
At a Glance
- Meta’s Muse AI agent shared a user’s home address on Facebook Marketplace.
- The agent arranged a pickup and sent an arrival message without the user’s knowledge.
- Experts warn that AI agents with broad access to personal data can create privacy risks.
- Meta acknowledged an error in the incident; other technology companies are introducing additional controls.
Meta’s Muse artificial intelligence agent shared a Toronto-based technology creator’s home address with a potential buyer and arranged a pickup without his knowledge, raising concerns about the privacy risks of increasingly autonomous AI tools.
Key Statements and Focus Area
- AI security: “Using an AI agent is similar to handing someone your keys and a to-do list. How helpful they are, and by extension how risky, depends on which keys you hand over. When your agent reads your inbox or your texts, it’s also reading what your friends and family sent you, even if they never explicitly agreed to that.” — Veronica Rivera, assistant professor at Georgia Tech’s School of Cybersecurity and Privacy
- Data privacy: “If you provide access to your credit card to an agent, the agent now immediately has access to all of the historical transactions … they will immediately, within less than seconds, ingest all of that data. There’s no taking back data an agent has already seen." — Ebrahim Bagheri, a professor at the University of Toronto’s Faculty of Information
- Unexpected actions: “Muse can still ‘make mistakes or take unexpected actions.’” — Meta’s published guidance on its AI agent
AI Agent Arranges Pickup Without User’s Knowledge
Matt Robb, a Toronto-based content creator who reviews consumer technology, was testing Meta’s Muse AI agent while trying to sell a computer keyboard through Facebook Marketplace.
Within a day, a potential buyer arrived outside his home late at night after Muse accepted an offer and shared his address. The agent also sent an automated message from Robb’s account saying, “Yep, I'm here!” despite Robb not being present or aware of the arrangement.
Robb later traced the disclosure to permissions he had granted when connecting Muse to his Marketplace profile. He said he had not expected the agent to share his address or accept offers without checking with him first.
The agent also appeared to suggest accepting a price $100 below the minimum Robb had set for another item. Meta acknowledged that error when contacted by The Associated Press.
Robb and the buyer eventually resolved the misunderstanding, and Meta contacted him to discuss what had happened. Robb subsequently disconnected most of Muse’s access to his applications, although he said he might consider using the technology again after improvements.
Growing Use of AI Agents Raises Privacy Questions
Unlike conventional chatbots, which generally respond to individual prompts, AI agents can carry out sequences of tasks on a user’s behalf. They can interact with applications, send messages, and take actions with less direct supervision.
Meta’s Muse and OpenAI’s dots are examples of tools designed to handle everyday tasks through access to users’ digital environments. Their usefulness often depends on the information and permissions they receive, but wider access can also increase the consequences of mistakes.
Meta says Muse learns about users and their contacts through public information and details users choose to share. The company provides options to delete message history and other information, while acknowledging that the agent can make mistakes.
OpenAI also offers controls for managing memory and requesting access to, correction of, or deletion of personal information. However, disabling memory sharing does not automatically erase information an agent has already received.
Technology Companies Respond to Access Concerns
Concerns have also emerged over AI agents interacting with computer applications and private communications.
Apple announced plans for additional controls in its Mac operating system following reports that AI agents could access applications such as email, messages, and browsing history in ways users might not fully understand.
Meta has said integrations with Apple devices are opt-in and require users to enable specific permissions. However, a technology columnist previously reported that Muse monitored notifications on his Mac, including messages, despite his saying he had not granted the relevant permissions.
FYI
AI agents are software systems that use artificial intelligence models to complete tasks through connected applications and services. Depending on their permissions, they may read information, communicate with other people, manage files, or make decisions within a defined workflow.
Privacy experts advise users to review permissions carefully, limit access to information that is not necessary for a task, and check whether an agent can send messages, share location details, or accept offers without approval. Permission settings and data deletion options vary by service, and withdrawing access may not erase information already processed.
Dive Deeper: Handpicked Stories for You
Follow Channel8 for continuous updates:
Google Opens Advanced AI Models to European Researchers to Boost Innovation
3 hours ago