Cybersecurity Research

Autonomous AI Tool Cairn Used in Campaign Against Semiconductor, Defense and Civil Society Targets

Unit 42 · 13 Nov 2026
Key Takeaway Assume attackers can now automate probing at speed, so prioritise patching exposed systems, enforcing multi-factor authentication and monitoring for unusual access.

Researchers at Unit 42 have described a campaign in which Chinese-speaking attackers used Cairn, a publicly available AI penetration testing framework, to run operations largely on its own. Between August and September 2026, the attackers carried out 25 operations against at least 19 organisations. They paired Cairn with the Chinese AI models Zhipu GLM and DeepSeek. Unauthorised access or data retrieval was achieved at 10 of the targeted organisations.

The attackers gave the tool a target and a goal, and Cairn then planned, executed and recorded results without further input. Each Cairn "agent worker" is driven by an AI model, and the workers share findings on a common "blackboard". Confirmed findings are stored as facts, while the next actions are queued as intents. Human operators can add "hints" at any time to steer the work if needed. The attackers ran a customised version of Cairn from a server in Hong Kong, with a staging node in the United States.

Unit 42 assesses with moderate confidence that this is China-nexus intelligence collection, which it tracks as CL-STA-1238. The campaign targeted three sectors: semiconductor, defense and civil society. This summary covers only the opening of the report, so the full details of the targeting and the compromised organisations are not included here.

Primary source @FBIView post ↗
Building or buying AI systems? Governing them under ISO 42001 ->

Summarised by CISO AI from Unit 42, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.