Meta has so far paid about $1.2 million in bug bounties for Muse, its personal AI agent that acts on a user's behalf across email, calendar, messaging and other accounts. According to SecurityWeek, the program ran privately from April and went public at the agent's September 8 launch. Of more than 630 reports, 64 were valid.
The largest award, $300,000, was paid four months before launch for a flaw that exposed conversations and tokens on internal virtual machines Meta staff used for testing. According to Meta's David Singleton, an authentication check trusted a header added by a local proxy, and after an infrastructure change requests from outside could carry it too. Meta took the service offline, fixed the bug and stopped treating the local machine as trusted.
$90,000 went to a malicious web page that, when the agent was asked to summarize it, mimicked status messages from its browsing tool and got it to take actions that should have been blocked; Meta added detection signals and changed how it handles suspicious tool output. Another $90,000 went to formatting that slipped one-time passcodes past Muse's email redaction filters, and $50,000 to an Android app flaw. Meta pays up to $300,000 per report.