Aurora ransomware operators leveraged SpaceX's AI Cursor Agent to accelerate exploitation activities against 10 victims between April 8 and May 26, 2026, according to a study published August 27 by Gambit Security's Threat Intelligence team. The attackers ran Claude Sonnet through Cursor Agent—a tool normally used by software developers to complete complex coding tasks independently—to assist with reconnaissance scanning, VPN client installation, and certificate attacks. While the agent didn't always succeed in its assigned objectives, the research shows how threat actors are continuously testing AI tools to accelerate and improve their campaigns.

The Aurora operators deployed Cursor Agent to help with post-compromise activities after already gaining credentials or an existing route into victim organizations. Commands given to the agent included gathering intelligence on the victim's environment, with some instructions as simple as "tell me what rights the user has" while others provided specific directions about which exploitation tool to use or followed a previously created attack plan. Tasks assigned to the agent ranged from enumerating the domain to report user privileges, using NetExec's BloodHound collector, and scanning internal subnets for hosts with Nmap or NetExec. Cursor Agent also received instructions to attempt NTLM relay attacks by coercing authentication with PetitPotam, Coerce Plus, and PrinterBug, running certificate attacks with Certipy, installing VPN clients or proxychains, then configuring and connecting to victims with supplied credentials or an existing SOCKS tunnel.

Most of the commands failed to reach the stated objective on the first try, leading to multiple refinements and changes to the commands and scripts used for each task, the researchers wrote. Some eventually succeeded in reaching the objective, while others failed and returned only a report of the attempts to the attacker. The study also observed Aurora operators rolling out a new Linux ransomware variant capable of targeting ESXi environments, using esxi_finder.py—a custom NetExec LDAP module—to scan for VMware ESXi hypervisors and vCenter servers inside victim networks. The variant encrypts virtual machine files and skips system volumes, keeping the hypervisor itself bootable so victims can read the ransom demand.

The use of AI coding agents in ransomware operations represents a shift in how attackers approach post-compromise activities, allowing them to automate reconnaissance and exploitation tasks that traditionally required manual execution by skilled operators. Gambit researchers also identified a second cluster of activity they attributed with medium confidence to an Aurora operator, in which eight victim organizations were targeted across Israel, Germany, Austria, Spain, the US, and Argentina. Aurora ransomware activity has been observed since April 2026, with the group operating a data leak site and targeting organizations across multiple countries. The deployment of specialized ESXi variants alongside AI-assisted tooling suggests Aurora is building a more sophisticated infrastructure to scale attacks against enterprise virtualization environments. Organizations relying on AI development tools may need to consider how these platforms could be repurposed by adversaries who already hold network access, particularly when those tools can execute terminal commands and modify system configurations autonomously.