Two developers, Peter James and Jonny L. Saunders, say they independently prompted Meta’s Muse into zipping up and sharing the full contents of its root filesystem, including Ubuntu system files, app templates, and internal documentation. Saunders posted on Mastodon that replicating James’ results was “extremely easy” and that Muse had “almost no prompt injection resistance.”
Meta denies that the incident represents a security breach. According to its announcement post, Muse runs in a persistent Linux virtual machine for each user.
Why it matters
The reports point to concerns about prompt injection resistance in AI tools that operate within a filesystem environment. The developers characterize the behavior as easy to reproduce, while Meta disputes that it constitutes a breach — a disagreement over how to interpret the exposure of system files and internal documentation.
Who should care
Developers working with AI agents that access underlying filesystems, and security researchers focused on prompt injection, may find the reported behavior relevant.