Human Hacks Marimo in Seconds Without AI

A recent analysis by Sysdig reveals that human attackers can execute complex intrusions with a speed previously associated with automated tools. In one documented case, an operator moved from an initial entry point to a secure server host in just eight seconds. This rapid sequence was achieved entirely through manual coding, without the assistance of artificial intelligence agents.

The incident highlights a significant shift in threat landscape dynamics. While industry narratives often suggest that AI lowers the barrier for cybercrime, this event demonstrates that individual skill remains a formidable force. The attacker leveraged a critical flaw in Marimo, a web-based notebook application, to gain full control of the system almost instantly after gaining initial access.

Manual Code Beats Automation

The operator wrote and debugged a custom Python toolkit by hand to exploit the vulnerability. This manual approach allowed the hacker to bypass detection traps that typically catch automated scripts. As reported by The Hacker News, this capability suggests that defenders cannot rely solely on blocking known AI-driven patterns to stay secure.

The attack exploited CVE-2026-39987, a severe remote code execution flaw affecting all versions of Marimo. The attacker used this gap to obtain an interactive shell, then retrieved credentials from Amazon Web Services Secrets Manager. Finally, they used a private key to authenticate to the target bastion host, completing the chain in under a minute.

Speed Reflects Deep Expertise

Sysdig researchers noted that the eight-second pivot is comparable to speeds seen in AI-assisted attacks. However, the human operator achieved this through deep technical knowledge rather than automated speed. They issued over 850 commands during a nine-hour session, all crafted manually within the compromised environment.

This methodology indicates a high level of tradecraft. The attacker avoided using publicly available offensive tools, instead building their own scripts on the fly. This level of customization makes it harder for security systems to identify and block the activity, as the code does not match known malicious signatures.

Other Campaigns Target Redis

Concurrent with the Marimo incident, Hunt.io disclosed a separate cryptomining campaign affecting over 3,500 Redis servers. This attack relies on missing authentication rather than specific software bugs. Attackers used the SLAVEOF command to deploy mining software, targeting a wide range of operating systems and Redis versions.

The Redis campaign highlights a different vector of risk, focusing on configuration errors rather than code flaws. While the Marimo incident showcases the speed of skilled humans, the Redis sweep demonstrates the scale of automated exploitation. Both events underscore the need for robust authentication and rapid patching across all digital infrastructure.