Meta employees testing its Muse AI agent reported privacy incidents, according to a Reuters investigation. Employees said the agent sent messages about topics they had not asked about, including career advice, childcare follow-up, products they had not bought, and bank account balances they did not recognize.
One employee said his Muse added an unprompted message to a fishing weather chat that read: "BTW, two bank updates, your Capital One balance is $2,722.58, your Charles Schwab balance is $5,899.57." Reuters reported that Muse also booked a non-refundable Airbnb for an employee who had not instructed it to do so.
Reuters said the employee concerns suggested Muse could be mixing up conversations between accounts and exposing private data. Muse also told another employee, who used the agent on her personal phone, about a Signal message. She said she had not given the agent access to Signal.
Reuters and the AI Now Institute reported that, in testing on an Android phone and a Mac, Muse used permissions granted during initial setup to pull message text. The setup included automatic access to Android push notifications, so the agent did not need explicit app access to read messages. The tests did not work on an iPhone.
Meta told Reuters the incidents were either isolated or the result of hallucinations. Reuters said it could not confirm this.
The source also cites earlier incidents. A cybersecurity professional reported a zero-day bug that made Muse hijackable. A journalist reported that Muse accessed his Apple Messages after he had denied that permission. The source says Apple then issued stronger privacy protections. The source also reports that banks have blocked agents over concerns, and that a user said Muse shared his address with a Facebook Marketplace buyer.
The source reports that Muse reportedly reached 10 million downloads in 40 days.