Meta Reports Muse AI Agent Bug Bounty Findings as Payouts Reach $1.2 Million
Summary
Meta says it paid about $1.2 million for Muse AI agent bug reports. A pre-launch flaw exposed data on internal test systems; Meta took the service offline, fixed the bug and described additional findings and defenses.
Key points
- Meta received more than 630 reports through the Muse bug bounty program; 64 were valid.
- A pre-launch authentication flaw let external requests carry a header trusted by Muse, exposing conversations and tokens on staff test VMs. Meta took the service offline and fixed the issue.
- A malicious page reportedly mimicked Muse browsing-tool status messages and induced actions; Meta says its updated detection catches the attack and 100 variants in testing.
- A report found that specific formatting could bypass Muse’s email redaction for one-time passcodes.
- An Android app flaw allowed a malicious app to pass a link to a privileged web view without host validation, potentially exposing its credentials.
- A Mac app flaw could let someone with access to the computer hijack auto-update to execute code with the app’s permissions.
Article Details
- Event Type
- Meta disclosed security findings from its Muse AI agent bug bounty program.
- Impact
- Meta paid roughly $1.2 million for valid reports. A pre-launch flaw exposed conversations and tokens on internal test VMs used by Meta staff. Other reports showed a malicious web page could induce Muse to perform actions, formatting could bypass one-time-passcode redaction, an Android app flaw could expose web-view credentials, and a Mac app flaw could allow code execution with the app’s permissions. Meta says it fixed the pre-launch flaw, took the service offline during remediation, and added detection signals and changes to how Muse handles suspicious tool output.