🗂️ The author asked Muse to archive the files it could access and send them to Google Drive. The resulting archive contained parts of the Linux system, documentation, agent logs, and SSH key files; it was about 2.7 GB compressed and 6.8 GB unpacked.
🔎 The author reported the findings to Meta and did not publish the archive, keys, or logs. They also said they could not confirm whether the keys were active and did not demonstrate a container escape. Meta marked the report “Not Applicable,” so the case alone does not prove an external compromise.
The story raises an important question about agents with persistent access: which files can they read, and where can they send them? Limiting permissions and reviewing connected integrations can reduce the risk of exposing internal data.
💡 Explanation in a nutshell#
An agent can only share files it is allowed to see and send. That is why its read permissions and export destinations need careful control, just like those of an app or user account.
More information at the link 👇
