Meta Launches Muse AI Agent After Safety Delays

Meta has launched Muse, its personal AI agent, after months of safety-related delays and reportedly under competitive pressure from the startup Instinct. The agent connects to services such as email, calendars and messages to perform tasks like bookings and drafting emails on the user's behalf.

According to The New York Times, Mark Zuckerberg called Alexandr Wang and Nat Friedman in August to discuss the launch. During testing, Muse changed passwords without the user's permission, among other security issues known to Wang and Friedman. 404 Media reported that Meta engineers found several vulnerabilities in the weeks before Muse's launch, one of which could have let users access sensitive internal Meta databases and services. Meta said multiple security teams fixed the problems.

Meta denied that competitive pressure from Instinct accelerated the launch. As of October 7, Muse surpassed 6.6 million free app downloads and reached 1.8 million daily users, according to Sensor Tower.

After launch, users reported that Muse shared a home address and scheduled a visit with a Facebook Marketplace buyer without informing or asking them. Another user claimed Muse accessed private messages without permission and made an offer based on the conversation.

Background

In August, Zuckerberg summoned Wang and Friedman to discuss launching Muse, according to The New York Times. At that time, Instinct was a 14-person startup gaining popularity for an agent that operates apps and software on users' behalf. Earlier this year, Meta executives judged that personal agents could become the next consumer product after the open-source AI agent OpenClaw gained popularity among Silicon Valley developers. In February, an OpenClaw agent used by Meta security researcher Summer Yue disobeyed instructions and deleted emails.