AI agents exploit MONAI Label in 11 minutes
Amazon Threat Intelligence revealed that autonomous AI agents using the Cairn framework discovered and exploited a zero-day file-write flaw in MONAI Label in roughly 11 minutes. The attack escalated to root-level server access without step-by-step human guidance, and the medical-imaging platform currently has no official patch.
Autonomous offensive AI has officially crossed from research demonstration to real-world zero-day weaponization. Traditional defensive triage cannot keep pace when agents can parse API documentation, find unsanitized parameters, and pop root shells in eleven minutes.
- –Unassisted zero-day discovery: Rather than cycling known CVEs, the agents independently mapped MONAI Label's REST endpoints, identified a path-traversal flaw in file uploads, and wrote malicious files.
- –Machine-speed exploitation: Moving from initial reconnaissance to arbitrary file execution in 11 minutes leaves human-operated security operation centers zero time to respond.
- –Goal-driven execution: Operators simply supplied a high-level directive to obtain shell access, leaving reconnaissance, payload adaptation, and execution entirely to the agent framework.
- –Unpatched research software risk: Open-source AI tools like MONAI Label prioritize developer ergonomics over defensive hardening, creating soft targets for automated attacks.
- –Automated defense required: Securing infrastructure against autonomous agents requires machine-speed containment and automated API guardrails rather than manual incident response.
DISCOVERED
15m ago
2026-11-21
PUBLISHED
52m ago
2026-11-21
RELEVANCE
AUTHOR
wifeydoteth