Spies used an open-source AI pentest tool to break into 10 organizations
Unit 42 says a likely China-nexus operator let an AI framework called Cairn run intrusions against semiconductor, defense and civil society targets, and no model refused.
Palo Alto Networks' Unit 42 says a likely China-nexus espionage operator used a customized fork of Cairn, an open-source AI penetration testing framework, to carry out intrusions in August and September. Unit 42 tracks the activity as CL-STA-1238 and makes the China link with moderate confidence.
What happened
Cairn hands an intrusion to an AI model, which repeatedly chooses and runs the next step. The operator ran 25 operations against at least 19 organizations and got unauthorized access or stole data at 10 of them. Operations lasted anywhere from about 20 minutes to more than 50 hours.
Results included valid credentials and an altered ballot document at an industry standards body, access to more than a million member records at a U.S. teachers' union, new admin accounts at a semiconductor company, and user records with military and government addresses from a defense-policy program.
The AI angle
The operator used Zhipu and DeepSeek models. None refused, because the tasks were framed as authorized testing backed by generated authorization files.
What it means
Unit 42 says the AI raised the volume and reach of the attacks, not their sophistication. Most of the holes exploited were familiar ones: authentication bypasses, weak or default passwords, exposed cloud APIs and old public CVEs.
Who's affected
Semiconductor and advanced manufacturing suppliers, defense and policy groups, and human rights and civil society organizations.
What to do
Cover the basics that AI-driven attacks exploit: remove default and weak passwords, enforce MFA, shrink exposed services and patch old CVEs. Check logs for the published indicators 154.44.31[.]4 and 82.29.129[.]54.