LLM agent observed in Marimo post‑exploitation (CVE‑2026‑39987)

Colleagues, a note: Sysdig detected use of an LLM agent following an RCE in Marimo.

What happened:
- Public Marimo instance compromised; cloud credentials stolen and an SSH key retrieved from AWS Secrets Manager.
- The key enabled eight parallel SSH sessions via a bastion and exfiltration of an internal PostgreSQL database within minutes.
- LLM indicators: adaptive commands, passing outputs between steps, and a planning comment in the command stream.

Why it matters: the agent increases attack adaptability and reduces effectiveness of static defenses.

Recommendations: upgrade Marimo to 0.23.0, hunt for public instances and rotate keys.

How do you assess the risk to your cloud environments?

#cybersecurity #cloud #LLM