i
News
News · 2026-11-04

Former Meta security chief calls Muse a “crazy” privacy risk

Dataist Dataist

Former Meta security chief Alex Stamos warned that giving the AI agent Muse access to personal messages, email and password managers is dangerous. Some users, he said, are effectively opening up their entire digital lives to the agent. Meta says Muse runs in an isolated virtual machine, but acknowledges that automated systems are imperfect. Researchers have already found vulnerabilities, and tech blogger Matt Robb said the agent revealed his home address to strangers.

Cover: Former Meta security chief calls Muse a “crazy” privacy risk

What data Muse can access

Muse can review years of private messages and pull information from an inbox, from old emails to former partners to passport numbers. Users also give it access to password managers. Stamos called this practice completely crazy.

Interest in AI assistants that handle everyday tasks on their own is growing beyond Meta, too. Instinct, a startup founded by a 23-year-old research scientist, raised $1 billion in a September funding round just a few months later.

Why agents create problems for banks

Financial companies are finding it harder to distinguish real users from their AI bots and suspicious activity. To act on a person’s behalf, agents bypass security measures. Stamos said Instinct had managed to evade several automated bot-detection systems, which, in his assessment, partly explains the company’s high valuation.

“These aren’t stolen credentials, but obviously bots are doing the interacting,” Stamos said on Alex Kantrowitz’s Big Technology Podcast.

Meta says Muse runs in a virtual machine isolated from other users’ agents. The company also warns that automated systems are imperfect and advises users to review Muse’s actions and grant only the access they are willing to give it.

Vulnerabilities and data exposure

In September, security researchers found vulnerabilities that could have given attackers access to sensitive personal data. Meta then strengthened its safeguards. Last month, researchers found other ways to expose such information: the risk arises when users share conversations through web links.

Tech blogger Matt Robb also claimed that Muse told strangers his home address and directed them to his home through his Facebook Marketplace account.

Dataist’s take: Muse requires access to data that lets the agent act on a user’s behalf, but also creates privacy risks for that user. Isolating the virtual machine does not eliminate those risks: Meta itself acknowledges that automated systems are imperfect, and researchers have already found vulnerabilities.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X