Unit 42: Chinese-speaking attackers ran 25 autonomous Cairn operations against chipmakers, defense and civil society

On November 12, 2026, Palo Alto Networks’ Unit 42 reported that Chinese-speaking attackers, tracked as CL-STA-1238, used Cairn, an open-source AI penetration-testing framework, to run 25 autonomous operations against at least 19 organizations between August and September 2026. “The attackers supplied a target and a goal, then Cairn planned, executed and recorded results on its own.” The operators ran a customized fork with an added skill pack, and powered it with Zhipu GLM 5.2 (13 operations), DeepSeek V4 Flash (8) and DeepSeek V4 Pro (4). Operations lasted from 20 minutes to more than 50 hours.

Targets included seven organizations in the semiconductor and advanced-manufacturing supply chain, three in defense and defense policy, and four in human rights and civil society, among them a Canadian human-rights body and a US teachers’ union. Unit 42 confirmed unauthorized access or data retrieval at 10 organizations. Outcomes included dozens of valid credentials at an industry standards body, a brute-forced production credential at a semiconductor company used to create admin accounts and a persistent API token, more than one million member records and about 31,000 CRM contacts at the teachers’ union, about 500 user records including .mil and .gov accounts at a defense-related target, and about 150,000 internal documents at an industrial-automation manufacturer. To get past model safeguards the attackers used a Chinese-language authorization preamble and an AUTHORIZATION.md file; neither model family refused.

Why it matters: earlier exposures of Cairn showed criminal clusters; this is an assessed espionage operation using the same open-source harness and cheap open-weight models against strategically chosen targets. Unit 42’s conclusion is the useful one for defenders: “Autonomous AI raises the volume and coverage of exploitation, but not its complexity.”

What it does not show: the China-nexus intelligence attribution is held with moderate confidence, resting on language artifacts, the choice of Chinese models and target selection. The weaknesses exploited were known ones such as authentication bypasses, weak credentials and old CVEs, and the report gives no per-operation success rate.

Sources

Last verified November 16, 2026