HomeNews
Share

Reading Signal chats, other people's purchases and bills: Meta Muse agent faced complaints

Meta employees are complaining that the company's new AI agent is reading their private messages and fabricating financial data

Yuliya Kotova

Yuliya Kotova

Metas AI agent, Muse, has faced complaints about privacy violations / Photo: Erlin Diah / Shutterstock.com

Meta's AI agent, Muse, has faced complaints about privacy violations / Photo: Erlin Diah / Shutterstock.com

Meta employees are concerned about security issues with the personal AI agent Muse, on which the company is banking in the hope of generating billions in revenue from artificial intelligence, Reuters reports. Since September, the AI agent has been downloaded more than 10 million times. In internal communications, Meta employees complain that the bot—designed to make users’ lives easier—has been accessing encrypted personal messages without authorization, fabricating purchase data, and sending messages apparently intended for to strangers, the agency reported.

What issues did Meta employees identify?

Reuters claims to have reviewed more than 20 internal complaints about Muse's unauthorized actions. Many of them were posted in a feedback group where Meta employees report identified issues to the Superintelligence team, which is developing the AI agent.

  • According to the complaints, Muse sent users messages about childcare requests they hadn't made, career advice they hadn't asked for, orders for products they hadn't placed, and bank account balances that didn't belong to them.

  • For example, in one instance, an AI agent added the following comment to a conversation about the weather and fishing for no apparent reason: “By the way, here are two updates on your bank accounts: your balance at Capital One is $2,722.58, and at Charles Schwab, it’s $5,899.57.”

Within the first 24 hours of its launch, Metas app featuring the AI agent Muse reached the top spot on the U.S. App Store. Photo: Robert Way / Shutterstock.com

Agent Muse in Action: How Far Will Meta Go with It in the AI Race?

  • In another instance, a Meta employee discovered that Muse on her personal phone had reminded her to reply to a message from her sister on the encrypted messaging app Signal. “This looks like a serious privacy violation. I definitely didn’t grant access to other apps, and I don’t remember being asked about it,” the employee wrote in an internal post. Her colleague replied that he, too, had encountered unwanted messages from Muse related to third-party apps and discovered that the AI agent automatically gains access to push notifications on Android phones.

  • Employees also reported other issues: Muse notified users about nonexistent transactions, overcharged them for items they had actually purchased (causing users to lose real money), and, in one instance, booked a non-refundable Airbnb reservation without permission.

  • In another instance, the AI agent unexpectedly suggested “writing to the seller” about a coffee table that the seller wasn’t interested in. When an employee asked Muse what this was about, the agent replied that it had “received some information about the pickup of a table that apparently does not belong to you.” The employee concluded that Muse had pulled up context related to another user and expressed concern about a possible data leak between accounts.

  • A representative of the Superintelligence team developing Muse reported in a feedback group in early November that the agent appears to be fabricating messages in response to built-in prompts to interact with the user between sessions. “Unfortunately, these hallucinations most often involve the most unpleasant topics, such as finances, security, and safety, which is terrible,” he noted.

Mark Zuckerberg sold Meta shares for the first time in a year / Photo: Meta

Mark Zuckerberg sold $21 million worth of Meta stock amid the buzz surrounding the AI agent Muse

How did Meta respond?

Meta spokesperson Daniel Roberts called the examples cited by Reuters “isolated incidents.” He stated that the company prioritizes security, conducts extensive internal testing, and continuously updates its products to improve security.

According to him, many of the controversial posts were “hallucinations”—that is, instances where the AI model makes up information—and this issue has already been resolved.

Reuters was unable to determine whether the AI agent Muse provided users with real data belonging to other people—such as bank account balances—or simply made up that information.

Context

Incidents involving Muse highlight the risks inherent in a new category of consumer AI agents that are entering the market and being positioned as the next big wave in the development of artificial intelligence, Reuters reports.

Amazon prohibits bots on its platform because they do not generate revenue from ad impressions / Photo: Thaspol Sangsee / Shutterstock.com

Amazon has blocked Meta's AI agent on its marketplace. What's the issue?

Meta is currently ahead of its competitors in this market. The Muse app has become one of the fastest-growing in history: it was downloaded over 700,000 times in the first 24 hours after its launch, and about 900,000 times in the first week. By comparison, ChatGPT’s debut week on iOS in 2023 resulted in 500,000 downloads. The app is being actively promoted among the 3.6 billion users of Facebook, Instagram, and WhatsApp. In addition, Muse has a high user retention rate: according to Meta, about a quarter of users who created accounts on the app return to it daily.

For AI agents to function effectively, they need broad access to smartphones and computers. However, they are also prone to making mistakes, exploiting little-known vulnerabilities, and achieving their goals in ways that can sometimes cause harm. For example, major AI developers OpenAI and Anthropic have reported instances where their AI agents unexpectedly hacked into other companies’ systems and independently compromised the systems designed to monitor their activities. Some banks, retailers, and other businesses have responded to these risks by blocking AI agents’ access to their systems.

This article was AI-translated and verified by a human editor

Share

Trending

Stock Screener
Buy
Sell
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
‌
Small Caps
Investment and Finance News