Meta has denied allegations that its AI agent, Muse, accesses private user messages without authorization. The controversy began after Inc. columnist Jason Aten reported that the AI read his messages. Meta's PR vice president, Andy Stone, stated that the Mac version of Muse requires users to manually enable full disk access and a message connector to function.
David Singleton, an executive at Meta's Reality Labs, explained that the macOS system architecture makes unauthorized access technically impossible. Users must navigate three separate permission layers, and the system requires a manual confirmation in the macOS settings menu. Singleton noted that even if a software bug existed, the OS security mechanisms would prevent the AI from bypassing these restrictions.
Aten claimed that Muse read his messages while full disk access was disabled, suggesting the AI might have parsed text from system notifications. Singleton disputed this, attributing the AI's response to an error in its internal processing. Meta maintains that the incident described by the journalist did not and could not occur under the current security framework.
The company faces ongoing scrutiny due to its history of data handling issues, including the 2018 Cambridge Analytica scandal. Recent reports also include a claim from YouTuber Matt Robb, who alleged that Muse leaked his home address during a Facebook Marketplace transaction. Meta is currently investigating this separate incident, which the user attributed to previously granted permissions.
