China hackers used AI agents to hit 2,250 companies at once
A China-linked group used AI agents to breach 2,250 Microsoft cloud tenants, gaining admin access in under three hours.
Drama · Source: Reuters
What happened
Microsoft caught a China-aligned hacking group using AI agents to attack thousands of companies globally. The group, tracked as Storm-0940, hijacked the ConnectWise Cloud Backup application to launch a massive supply chain attack. They used this trusted access to target over 2,250 Microsoft 365 cloud environments. The threat actors have been active since at least 2021 and previously used a botnet called Quad7.
The hackers armed themselves with ARTEX. This is a Go-based agentic AI penetration testing framework. They weaponized it to automate reconnaissance, password spraying, and credential theft at machine speed. The AI agents hammered targets with up to 2,900 login attempts per minute. Between October 10 and 14, Microsoft captured over 151,000 failed logon events. By October 15, the agents gained access to ConnectWise app identities and bypassed interactive user sign-ins entirely.
The speed of the compromise was brutal. In several cases, the AI agents went from initial targeting to full privileged access in less than three hours. Once inside, the agents modified credentials to create persistent backdoors. Security researchers found that the ARTEX deployment used models like DeepSeek, GLM, and Grok. The prompts actually directed the AI models to divide investigation tasks and review each other's findings. They hunted for bulk personal data, payment gateway API keys, and unauthorized administrative privileges.
Key facts
- 2,250 — Microsoft 365 tenants targeted by the AI agents
- 2,900 — Sign-in attempts per minute at peak activity
- 3 hours — Time from initial targeting to privileged cloud access
- 151,000 — Failed logon events captured over four days
Why it matters
The timeline for detecting a network breach just collapsed completely. Human security teams cannot respond to attacks moving at this machine speed. If an AI agent can map your network, divide tasks among subagents, and escalate privileges in under three hours, traditional manual incident response is dead. You need automated defenses to fight automated attacks. The blast radius of supply chain attacks is also expanding because AI agents can exploit trusted connections instantly across thousands of tenants.
Open source security tools are becoming weapons of mass compromise. ARTEX was originally built for penetration testing and was freely available. Now it is a fully autonomous attack framework used by state-sponsored actors. As developers build better agentic workflows, hackers will adopt them immediately to scale their operations without adding human headcount. The fact that these AI models are cross-checking each other's work shows a terrifying leap in autonomous hacking capabilities.
For builders
Audit third-party application permissions immediately
The hackers bypassed interactive logins by exploiting permissions already granted to ConnectWise services. Review all OAuth apps connected to your cloud infrastructure. Revoke access for anything you do not actively use, or you will pay the price for a vendor breach.
Prepare for machine-speed credential attacks
Attackers are hitting endpoints at 2,900 attempts per minute. Your rate limiting and anomaly detection must trigger instantly. If you rely on human alerts to stop a password spraying campaign, your company loses.
Watch for persistent backdoor credentials
The AI agents modified application credentials to maintain access even after the initial breach was closed. Security teams must look beyond the initial entry point. You have to audit identity and group changes after any incident to ensure the attackers did not leave a backdoor.
My take
Cyber warfare is no longer human versus human. It is AI versus AI, and if state-sponsored hackers are deploying autonomous agents to breach networks in three hours, your defense cannot rely on a tired security analyst looking at logs. I build AI tools every day, and seeing them weaponized like this proves we need agentic security just to survive.
Original reporting: Reuters. This is my rewrite and opinion.