Security researchers have revealed a cyberattack in which a skilled human operator exploited a critical Marimo remote code execution vulnerability and reached an SSH bastion host within eight seconds after gaining initial access. According to findings from cloud security company Sysdig, the incident demonstrates that experienced attackers can still execute highly efficient attack chains without relying on artificial intelligence tools. The research highlights how quickly attackers can move after exploiting vulnerabilities, as AI continues to reduce the time between vulnerability discovery and exploitation while making advanced techniques more accessible.
The attack involved exploitation of CVE-2026-39987, a pre-authenticated remote code execution vulnerability affecting all versions of Marimo. The flaw, which carries a CVSS score of 9.3, was actively exploited shortly after public disclosure. According to Sysdig Threat Research Team findings, the attacker moved from a vulnerable Marimo notebook environment to an SSH bastion host in just eight seconds by using a custom Python toolkit developed manually during the operation. Researchers noted that the activity resembled the speed commonly associated with AI assisted attacks, but the attacker achieved the results through technical expertise and manually created tools. After obtaining access through the vulnerability, the threat actor established a full interactive shell and executed a credential pivot process that involved accessing AWS Secrets Manager using credentials collected from the compromised instance. The attacker then retrieved a private key and used it to authenticate with an SSH bastion host.
Sysdig detailed the timeline of the activity, showing that a fresh WebSocket connection was established at 18:57:22, followed by a lookup against stored application credentials at 18:57:26 that returned the harvested AWS key. By 18:57:30, SSH authentication activity was detected at the bastion host. The complete session continued for approximately nine hours, during which the attacker executed more than 850 interactive commands. Researchers stated that the operator did not use recognizable publicly available offensive tools and instead developed scripts during the session. The attack chain involved a backgrounded Python3 process that retrieved credentials, obtained an SSH key from Secrets Manager, stored it locally, and connected to the bastion system. Sysdig noted that while AI may improve the efficiency of cyber operations by increasing attack speed and reducing repetitive manual tasks, skilled human operators remain capable of building custom techniques and avoiding security detection methods.
Additional research from Hunt.io has also highlighted a separate cryptomining campaign targeting thousands of Redis servers. The campaign reportedly compromised 3,562 Redis servers following broad internet scanning activity focused on potential Redis hosts operating on port 6379. Researchers identified multiple attack pipelines, including WordPress target discovery, AOF based SSH authorized keys injection, and Lua sandbox escape attempts. The primary method used in successful compromises involved the SLAVEOF command to introduce attacker controlled content into Redis servers, leading to the deployment of an XMRig cryptocurrency miner. Hunt.io reported that affected systems ranged from Redis versions 2.8.17 through 7.2.0 and included Linux environments from outdated RHEL and CentOS versions to current Ubuntu kernels. Researchers suggested that missing authentication was likely the main weakness rather than a specific software version issue.
The investigation also found that while several techniques were tested against different services, Redis rogue replication was the only method that achieved large scale success. SSH key injection and MongoDB sandbox escape attempts did not produce confirmed results across thousands of tests, while a WordPress credential to webshell chain was identified but not verified at scale. Separately, researchers have also linked a single operator to Operation CameraSwarm, a campaign that compromised more than 14,000 Dahua IP cameras through brute force attempts, authentication bypass vulnerabilities including CVE-2021-33044 and CVE-2021-33045, and peer to peer relay techniques. These incidents highlight the continued need for organizations to maintain strong security controls, apply timely patches, monitor exposed services, and strengthen cloud and infrastructure defenses against rapidly evolving attack methods.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
