What the buyer reportedly received
In an Associated Press report, Matt Robb describes his Meta Muse AI agent sharing his home address with a Marketplace buyer. This is Robb’s account as reported by AP; it is not a claim that every AI agent or every user experienced the same behavior.
Why an “always allow” permission matters
Agents can act across conversations and services when users grant them access. A broad permission may save time, but it also reduces the chance to review a particular message before it goes out. The practical question is which information the agent can use, to whom it can send it, and whether confirmation is required for sensitive actions.
What Meta acknowledged—and what remains unclear
Meta acknowledged an error involving a separate below-minimum price suggestion. That acknowledgment should not be read as confirmation of every detail of the address-sharing incident. The AP account illustrates a privacy risk, while the exact sequence and system behavior should be described with that attribution and uncertainty intact.
How to limit the risk—and where to listen
Review an agent’s permissions, narrow any “always allow” access, avoid sharing sensitive data unless necessary, and require approval before messages containing personal information are sent. Controls differ by product. Watch the Daily AI Chat Short or listen to the full podcast episode. Source: Wyatte Grantham-Philips, Associated Press, October 19, 2026.
