Some users are giving Muse their entire password vaults, according to Alex Stamos, Meta’s former security chief, who discussed this on Alex Kantrowitz’s “Big Technology Podcast” and described the idea as “totally insane.” A possible data breach isn’t the only concern. With those credentials, the AI agent can act across the services that make up someone’s digital life.
Stamos discussed the risks on the podcast, as Business Insider reports. Meta has encouraged people this year to give Muse their email login credentials and credit card information. Depicted as a furry avatar, the agent can search years of personal instant messages and access entire email inboxes, potentially finding everything from forgotten conversations with exes to passport numbers. Adding a password vault gives Muse still broader access, effectively handing it control of a user’s digital life.
Meta’s explanation of Muse’s safeguards says the agent runs inside a virtual machine that is “isolated so no one else’s agent can access it.” Yet the company acknowledges that “no automated system is perfect,” advising people to “review Muse’s activity and only grant the access you’re comfortable with.” Isolation offers a safeguard, rather than an assurance that everything Muse accesses or does will stay private.
Researchers’ discoveries illustrate the limits of those assurances. In September, The Information reported on vulnerabilities researchers had identified that could allow hackers to access sensitive personal data. Meta subsequently bolstered its security measures and expanded its safety warning, according to Reuters.
Last month brought additional findings: researchers identified ways attackers could expose sensitive information when people used web links to share conversations, Ars Technica reported. That creates a separate route for data to leak. Information an agent can access may be exposed without anyone stealing login credentials, because the software’s conversation-sharing features can themselves introduce a vulnerability.
Tech YouTuber Matt Robb has also alleged that Muse gave strangers his address. He also claimed that Muse sent strangers to his home via his Facebook Marketplace account. His description illustrates the possible consequences of allowing software to operate through a personal account. This remains Robb’s claim, though; it hasn’t been independently confirmed here.
Muse isn’t alone in seeking that kind of access. Instinct, the viral personal AI assistant startup, is developing agents that can autonomously complete day-to-day tasks. A 23-year-old research scientist founded the business just months ago, and it raised $1 billion during a September funding round.
The source article describes a major headache for the financial world: distinguishing suspicious activity from legitimate users and their AI bots. It also characterizes the agents’ work on users’ behalf as requiring them to break security measures by definition. Stamos said the credentials aren’t stolen, but the agents are clearly bots interacting with services.
“They’re not stolen credentials, but are clearly bots doing interaction,” Stamos said. “And one of the reasons Instinct has a ridiculous valuation from my perspective, but one of the things they did that made them popular is they actually figured out ways to defeat a bunch of the automated bot detection mechanisms from these companies.”
The source article considers Stamos’s caution warranted, particularly given what it calls Meta’s abysmal history of safeguarding personal data. Whatever an AI assistant’s practical value, there’s a more pressing question about how much authority people hand over. Giving Muse access to email, credit card information or a password vault lets it do more for a user. But that access also leaves a larger amount of sensitive information exposed if its safeguards or sharing features break down.
