On Tuesday, Meta Platforms, Inc. (NASDAQ:META) launched Muse, an AI agent designed to autonomously manage tasks across users’ digital lives. However, internal testing has reportedly uncovered security vulnerabilities and reliability problems.
Meta Muse AI Agent Can Send Emails, Book Travel
Muse, known internally as Hatch, is designed to go beyond traditional chatbots by taking actions on users’ behalf, including sending emails, booking travel and selling items.
The AI agent is central to CEO Mark Zuckerberg’s vision of "personal superintelligence" for Meta’s billions of users.
The assistant can connect to apps and services involving email, calendars, payments, health, shopping and smart-home devices. Users decide which apps Muse can access and can revoke those permissions.
The AI is initially available in the U.S. through a dedicated app and WhatsApp. Meta plans to bring it to its smart glasses soon. A basic version is free, while heavier users can pay $20 or $100 per month, Reuters reported.
Internal Tests Raise Muse Privacy and Security Concerns
Meta delayed Muse’s initial April launch to improve its security. Vishal Shah, Meta’s vice president of AI products, told the publication that the company believes the additional work helped it "cross the threshold" needed to release the product.
"It is impossible to say that there is never going to be a mistake," Shah said, while arguing that the system was designed to be "as safe, as secure, as private as we can possibly make it."
However, internal employee testing reportedly produced mixed results.
One test allegedly found Muse bypassing safeguards and exposing private iCloud photos after being asked to identify toys in pictures from a child’s birthday party.
Another employee reported "many failure modes that made it unreliable" while using Muse to monitor tickets and other fast-selling items.
Meta CTO Andrew Bosworth also reportedly experienced repeated login problems.
Meta did not immediately respond to Benzinga’s request for comment.
Meta Adds AI Safety Layer
Muse operates through its own virtual machine, allowing it to perform tasks in the background. Meta also uses a separate monitoring agent that reviews planned actions and can require user authorization for certain tasks.
Users can opt out of having their interactions used to train Meta’s AI models, while an encrypted version is planned for later this year.
The launch comes as other AI companies face similar challenges with autonomous agents behaving unpredictably, highlighting the growing tension between AI agents’ usefulness and the risks of giving them access to real-world data and accounts.
Price Action: Meta Platforms closed at $613.48 on Tuesday, down 0.53%, while shares were up 0.18% to $614.60 in after-hours trading, according to Benzinga Pro.
According to Benzinga Edge Stock Rankings, Meta ranks in the 79th percentile for growth, although its shares have underperformed over the long term while delivering positive returns in the short and medium term.
Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.
See More: Top Growth Stocks
Photo Courtesy: 24K-Production on Shutterstock.com
Login to comment