Human Attacker Exploits Marimo RCE at Machine Speed

• •CVE-2026-39987 exploited, allowing rapid SSH access.
• •Attacker executed over 850 commands in a nine-hour session.
• •Detection strategies must differentiate between human and AI-driven attacks.

A human attacker exploited CVE-2026-39987, a pre-authentication remote code execution vulnerability in Marimo notebooks, achieving a rapid transition from an open WebSocket to SSH access in just eight seconds. The attacker utilized a hand-rolled Python toolkit, bypassing detection mechanisms designed for AI-driven attacks. Over a nine-hour session, they executed more than 850 commands without using publicly available offensive tools. This incident highlights the potential for skilled human operators to match or exceed the speed of AI-assisted attacks. The vulnerability affects Marimo versions up to 0.20.4 and has been on CISA's Known Exploited Vulnerabilities catalog since April 2026. Sysdig's Threat Research Team documented the attack and provided recommendations for mitigating the risk. The incident emphasizes the need for robust detection strategies that account for both human and AI-driven threats.