A sophisticated intrusion campaign leveraging the CyberStrikeAI platform has been identified targeting Amazon Web Services (AWS) environments. Between September 30 and October 16, 2026, an attacker utilized a single server to navigate four distinct AWS accounts, employing eight separate identities to execute a series of malicious actions. The campaign is notable for its use of machine-speed automation, which allowed the attacker to perform 56 AWS API calls in just 141 milliseconds, effectively blending in with legitimate administrative traffic.
The activity originated from an IP address identified as 43[.]240[.]29[.]11. Researchers have attributed this infrastructure to CyberStrikeAI, an open-source framework that integrates large language model (LLM) agents with offensive security tooling. While the platform has previously been linked to attacks against Fortinet edge devices, this incident marks the first documented instance of the tooling being repurposed to target cloud infrastructure.
The intrusion unfolded in four distinct phases. On September 30, the attacker began by validating stolen AWS credentials. Shortly thereafter, the actor moved to harvest data from AWS Secrets Manager and retrieved EC2 user data. Between October 1 and October 2, the attacker proceeded to decrypt AWS Systems Manager (SSM) parameters and execute remote commands via the SSM SendCommand function. Following this initial burst of activity, the attacker went dormant for 14 days before returning on October 16.
The primary challenge in detecting this campaign was the attacker's ability to operate within the noise of standard cloud operations. Because the API calls mirrored routine administrative tasks, individual actions did not trigger high-severity alerts. Security researchers noted that six separate detection rules were triggered during the intrusion, but each rule only identified a small, isolated portion of the overall activity. The 14-day gap between the initial phase and the return of the attacker further complicated detection, as standard correlation windows in security information and event management (SIEM) systems are typically not configured to link events separated by such a long duration.
The security team investigating the incident confirmed that the attacker successfully utilized eight different identities to maintain persistence and move laterally across the four compromised accounts. While the speed and nature of the API calls strongly suggest the use of AI-driven automation, researchers have clarified that they cannot definitively prove an LLM agent was making real-time tactical decisions. Furthermore, the specific method used by the attacker to obtain the initial set of credentials remains unverified.
The incident highlights a significant shift in how automated offensive tools are being deployed against cloud environments. By executing high-frequency API calls, the CyberStrikeAI server was able to minimize its footprint and avoid triggering automated containment protocols. The use of multiple identities across multiple accounts suggests a deliberate effort to evade detection by spreading the malicious activity across a wider surface area.
Organizations are advised to review their AWS CloudTrail logs for the identified indicators of compromise, particularly focusing on rapid sequences of API calls that involve Secrets Manager, SSM parameters, and EC2 user data. Because the attacker relied on stolen credentials, the incident underscores the necessity of robust identity and access management, including the implementation of automated containment workflows that can respond to suspicious patterns even when individual events appear benign.
In response to the findings, security analysts emphasized that the primary danger lies in the inability of traditional, static detection rules to account for long-term, low-and-slow persistence. The 14-day gap demonstrated that attackers are increasingly aware of the limitations of fixed-window correlation. Future defense strategies must account for these extended timelines to effectively identify and neutralize automated threats that leverage AI-assisted infrastructure.




