ARTEX Explained: How a China-Linked Group Used an AI Agent to Target Thousands of Microsoft 365 Tenants
AI is changing how cybersecurity teams investigate threats, but attackers are also finding ways to use automation to increase the speed and scale of their operations.
Microsoft disclosed on Wednesday, October 21, 2026, that a China state-aligned actor it tracks as Storm-0940 was behind a recent compromise involving the ConnectWise Cloud Backup application.
The notable detail was the reported use of ARTEX, a Go-based agentic AI penetration testing framework that Microsoft said helped the group accelerate scanning, password spraying and the abuse of OAuth applications against downstream victims.
ARTEX had reportedly been made closed-source following earlier misuse against South Korean financial institutions.
Microsoft said the activity had been running at machine speed against hundreds of organisations worldwide since September 2026. The affected sectors included aerospace and defence, automotive, chemicals, energy, financial services, government, legal services, manufacturing, media, semiconductors, software and think tanks.
The incident highlights a growing concern for businesses: automated attacks can carry out repetitive security testing and account attacks far faster than human defenders can manually investigate every alert.
What Is ARTEX?
ARTEX is described in the supplied reporting as a Go-based agentic AI penetration testing framework.
Agentic AI refers to systems that can carry out sequences of tasks with some degree of autonomy rather than simply responding to one instruction at a time. In a security context, such a framework may help coordinate activities such as scanning, testing credentials and interacting with applications.
Microsoft said the group used ARTEX to accelerate several activities, including password spraying and abuse of OAuth applications.
The significance is not simply that an attacker used an AI-related tool. It is that automation can help coordinate multiple steps of an intrusion and repeat them at a pace that makes detection and response more difficult.
The framework’s earlier reported misuse against South Korean financial institutions also illustrates why dual-use cybersecurity tools can become a concern when their capabilities are applied against systems without authorisation.
How the Microsoft 365 Campaign Worked
According to the supplied reporting, stolen ConnectWise credentials gave the attackers an initial foothold.
From there, the group used automated activity to try passwords across accounts and abuse authentication applications and OAuth consent flows to reach privileged cloud access.
Microsoft’s account of the campaign points to a familiar set of security weaknesses: compromised third-party credentials, weak or reused passwords, and excessive trust in applications connected to a cloud environment.
The use of an automated agent can make these techniques more efficient. Rather than relying on a person to test each account or repeat each step, an automated framework can distribute tasks, run repeated attempts and adjust its activity as it proceeds.
Tom’s Guide, summarising Microsoft’s disclosure, reported approximately 2,900 sign-in attempts per minute and around 2,250 corporate tenants reached within hours.
Those two figures come from that outlet’s write-up and should be verified against Microsoft’s original disclosure before republication as confirmed Microsoft figures.
Why Password Spraying Remains a Serious Risk
Password spraying is a technique in which an attacker tries a small number of commonly used passwords against many accounts rather than repeatedly testing numerous passwords against one account.
This approach can be used to target organisations where employees reuse weak passwords or where account protection is inconsistent.
When the activity is automated, attackers may be able to test large numbers of accounts quickly. That does not mean every attempt will succeed, but it can create a substantial volume of suspicious sign-in activity.
Organisations should therefore avoid relying only on password complexity policies. They also need unique credentials, effective multifactor authentication, sensible sign-in restrictions and monitoring that can identify unusual patterns across accounts.
A password manager can help ordinary users generate and maintain unique passwords, reducing the risk that one compromised password will expose multiple accounts.
How OAuth Abuse Can Lead to Cloud Access
OAuth is a framework used to allow applications to obtain authorised access to services without requiring users to share their passwords directly with every connected application.
In Microsoft 365 environments, third-party applications may receive permissions that allow them to access certain organisational resources.
That functionality is useful, but it also creates a security risk when users or administrators approve permissions that are unnecessary, excessive or malicious.
The supplied reporting says the attackers abused OAuth application and consent flows as part of their effort to obtain privileged cloud access.
For organisations, the practical lesson is to review the applications connected to their Microsoft 365 tenants. Security teams should identify which applications have been granted access, what permissions they hold and whether each permission is still required.
Unknown applications and permissions that cannot be justified should be investigated and removed where appropriate.
How Businesses Can Protect Their Microsoft 365 Tenants
Businesses should take a layered approach to defending against automated attacks.
1. Use phishing-resistant multifactor authentication
Hardware security keys and passkeys can provide phishing-resistant authentication, helping protect accounts against common credential theft techniques.
Where possible, organisations should prioritise strong authentication for administrators and other accounts with access to sensitive systems.
2. Eliminate password reuse
Employees should use unique passwords for business accounts. Password managers can help generate and store strong credentials without requiring people to remember every password.
3. Audit OAuth application permissions
Review applications authorised to access the Microsoft 365 tenant. Identify unnecessary integrations, excessive permissions and applications that no longer have a legitimate business purpose.
4. Review third-party vendor access
Businesses often depend on external IT providers and backup platforms. Ask vendors with administrative access how they secure their accounts, manage credentials and protect their own systems.
ConnectWise customers should pay particular attention to Microsoft’s guidance concerning the backup application involved in the disclosure.
5. Monitor sign-in activity
Look for unusual authentication patterns, repeated password attempts, suspicious application consent activity and unexpected privileged access.
Because automated activity can generate large volumes of events, monitoring should help defenders identify patterns rather than depend entirely on manually reviewing individual alerts.
6. Limit administrative privileges
Give users and applications only the access they need. Reducing unnecessary privileges can limit the potential impact of compromised credentials or malicious application permissions.
7. Review incident response procedures
Organisations should be prepared to revoke compromised credentials, investigate suspicious application access and remove unauthorised permissions quickly.
AI Is Changing Both Cyberattacks and Cyber Defence
The ARTEX campaign forms part of a wider pattern in which AI and automation are increasingly relevant to cybersecurity.
The same week saw AI-assisted security research associated with the PHP vulnerabilities and a compromised AI development package distributed through LiteLLM.
These incidents involve different issues and should not be treated as a single coordinated campaign. However, together they illustrate that automation can appear on both sides of the security equation.
Security researchers may use AI models to help identify software weaknesses, while attackers may use automated agents to accelerate reconnaissance and credential attacks.
For defenders, the challenge is to improve the speed of detection and response while strengthening the controls that prevent common attack techniques from succeeding.
