Recent security research has revealed that the China-linked cyber threat group known as Storm-0940 has utilized the agentic AI-powered penetration testing framework called ARTEX to execute large-scale attacks targeting hundreds of organizations worldwide across various industrial sectors. In several instances, these automated attacks resulted in the attacker gaining privileged access to victims' cloud infrastructures, with the time from initial targeting to gaining privileged access being as little as three hours in some cases.
Technical data indicated that the malicious activity was directly linked to the recently announced compromise of the multi-tenant application supply chain of ConnectWise. These developments demonstrate how threat groups have begun integrating agentic AI systems into traditional offensive security workflows to significantly scale their operations, overcoming human limitations by working continuously around the clock and attempting to breach multiple vectors simultaneously.
Attack mechanism and execution speed via the ARTEX framework
The ARTEX framework is a multi-agent system based on large language models, written in Go with a Next.js-based user interface. It can execute fully autonomous penetration testing, from the reconnaissance phase to data exfiltration, with minimal human intervention. In the ConnectWise incident, the Storm-0940 group leveraged this framework to accelerate scanning and reconnaissance operations, password spraying, and application enumeration via the OAuth API.
- Password spraying attacks: The attackers' infrastructure executed an automated campaign targeting 327 accounts across 8 organizations to access Microsoft 365 and Exchange Online, recording thousands of failed authentication attempts at record rates reaching 2,900 failures per minute.
- Exploitation of application credentials: After gaining control of multiple cloud application identities belonging to ConnectWise, the AI agents used the same IP address range to authenticate into customer environments without the need for human interaction.
- Post-exploitation activities: This included reconnaissance via Microsoft Graph and modifications to application credentials to ensure persistence and change privileged roles and permissions.
What is practically changing for organizations?
These attacks represent a qualitative shift in the speed and scale of cyber threats; relying on automated agents drastically reduces the time available for defenders to detect, respond to, and counter an attack compared to traditional manual methods. Although the mere presence of the affected application or the occurrence of an authentication does not necessarily mean customer data has been leaked, modifications to service credentials can leave active backdoors even after the original application has been remediated.
Mitigation and technical protection guidelines
Security authorities have recommended taking immediate steps to contain risks and secure affected infrastructures through the following measures:
- Keep connections for affected ConnectWise applications disabled until official restoration guidance is issued by the provider.
- Refrain from deleting enterprise applications as an initial response, and focus on preserving configuration logs and forensic evidence.
- Review service principal identities and the four application IDs associated with the incident in every Microsoft 365 tenant.
- Examine Microsoft Entra audit logs and Microsoft Purview logs for any unusual additions of credentials or privileged role permissions.