Amazon Says AI Agents Exploited MONAI Label Zero-Day for Root Access

Summary
Amazon says autonomous AI agents exploited a path-traversal flaw in MONAI Label 0.8.5 and earlier, gaining root access on one university server. A source-code fix is merged, but no patched release is available.
Key points
- CVE-2026-133968 affects MONAI Label 0.8.5 and earlier; in the default setup, which has no login enabled, an unauthenticated attacker can write files outside the server's storage folder.
- Amazon says agents reached the file-writing stage in about 11 minutes; root access came later, in a separate session, on one university server.
- The agents reportedly planted a file that Python loads at startup, then triggered a training feature that launched a new Python process and connected back as root.
- Amazon reconstructed the activity after finding the operator's passwordless Cairn control panel exposed online, revealing agent prompts, task settings, and run history.
- Amazon assesses a China-linked actor with moderate confidence based on language, time-zone settings, and target selection; it identified no named group or government involvement.
- Amazon's account has not been independently confirmed. It says a second university network was likely compromised but that remains unconfirmed; it found no evidence that patient records were taken or clinical systems were affected.
- Maintainers merged a source-code fix on October 19, but no patched release is available; the newest release remains 0.8.5. The article advises cutting off exposure and limiting what a compromised server can reach.
Article Details
- Vulnerability Types
- Path traversal
- Severity
- Critical (9.8/10)
- Affected Versions
- MONAI Label 0.8.5 and earlier
- Exploitation Status
- reported
- Exploit Availability
- unknown
- Patch Status
- partial
- Workarounds
- Cut off network exposure to MONAI Label servers.
- Limit what a compromised server could reach.
MITRE ATT&CK
T1059.006 · PythonA training feature started a Python process that loaded the planted file and connected back to the attacker.T1190 · Exploit Public-Facing ApplicationThe agents reportedly exploited the network-reachable MONAI Label server's path-traversal flaw to write files outside its storage folder.
