Unit 42 Details Cairn AI Attacks Across 25 Operations Targeting 19 Organizations

Summary
Unit 42 reports that attackers used Cairn with Chinese AI models in 25 operations against at least 19 organizations; its investigation found unauthorized access or data retrieval at 10. Attribution to China-nexus intelligence collection is assessed with moderate with
Key points
- Unit 42 says attackers used a customized Cairn framework in 25 operations against at least 19 organizations from August to September 2026.
- Operators supplied targets and goals; Cairn autonomously mapped attack surfaces, broke goals into tasks, and repeated Reason and Explore phases until stopping or meeting its goal.
- The operations used Zhipu GLM 5.2 and DeepSeek V4 models, plus a custom “AIScan Runner Skills” pack with scanning and exploitation capabilities.
- Unit 42 reports that its investigation found unauthorized access or data retrieval at 10 organizations, including access to credentials and personal records, persistent administrator accounts, and modification of a live standards-ballot document.
- Targets included semiconductor and advanced-manufacturing organizations, defense and policy entities, and civil-society groups. The attackers sometimes supplied environment-specific information, including VPN details and shared passwords.
- Unit 42 assesses with moderate confidence that the activity was China-nexus intelligence collection, citing Chinese-language artifacts, tooling, and target selection.
- The article says the campaign exploited known weaknesses such as authentication bypasses, weak credentials, exposed cloud APIs, and dated public CVEs; it recommends identity hygiene, attack-surface reduction, and patching.
Article Details
- Attack Vectors
- Attackers used a customized Cairn fork to autonomously map targets, plan tasks, and carry out web application and service exploitation.
- The attackers ran Cairn with Zhipu GLM 5.2 and DeepSeek V4 Flash/Pro, and loaded the attack-specific AIScan Runner Skills pack.
- Weak-password spraying against an internet-facing application yielded valid credentials; brute force against a production credential enabled access to a partner-integration application.
- The activity exploited authentication bypasses, insecure direct object references, mass-assignment privilege escalation, weak or default credentials, exposed cloud APIs, and dated public CVEs.
- Other reported methods included unauthenticated account creation or registration bypasses, unauthorized access to non-public records, and overwriting a live standards-ballot document through an unauthenticated cloud-storage write.
- Defensive Notes
- The article recommends identity hygiene, attack-surface reduction, and software patching.
- For the specific activity, Palo Alto Networks states that Advanced URL Filtering and Advanced IP Defense identify known malicious domains and URLs associated with it.
- The article describes Cortex XDR and XSIAM as intended to detect and prevent known and unknown malware execution, and Cortex Agentic Endpoint Security as providing visibility and controls for AI artifacts and agent activity.
Indicators of compromise
| Type | Indicator | Context |
|---|---|---|
| IPV4 | 154[.]44[.]31[.]4 | Attacker-operated Cairn hosting server used to run the customized framework. |
| IPV4 | 82[.]29[.]129[.]54 | Attacker-operated staging and attack-egress server. |
MITRE ATT&CK
T1078 · Valid AccountsAttackers used valid credentials to access application data and accounts.T1110.001 · Password GuessingAttackers brute-forced a production credential for a partner-integration application.T1110.003 · Password SprayingAttackers sprayed weak passwords against an internet-facing application and obtained valid credentials.T1136 · Create AccountAttackers created unauthorized administrator accounts in a production system.T1190 · Exploit Public-Facing ApplicationThe operations exploited internet-facing applications and backend registration endpoints to gain unauthorized access.T1213 · Data from Information RepositoriesAttackers accessed non-public internal directory databases and retrieved application data, records, and documents.T1565.001 · Stored Data ManipulationAttackers overwrote a live industry standards-ballot document through an unauthenticated cloud-storage write.T1595 · Active ScanningCairn autonomously mapped target assets after attackers supplied organizations' root domains.
Threat Actors
Vendors
Products
Tools
AIScan Runner SkillsEach operation also loaded a skill pack named “AIScan Runner Skills,” providing the model with scanning and exploitation capabilities.CairnFrom August–September 2026, Chinese-speaking attackers used an open-source AI penetration testing framework called Cairn to autonomously conduct 25 operations against at least 19 organizations.
Countries
CanadaHuman rights and civil society: Four organizations including a Canadian human-rights body, a U.S.ChinaWe assess with moderate confidence that this activity represents China-nexus intelligence collection, which we currently track as CL-STA-1238.Hong KongAttackers ran a customized fork of Cairn from a server hosted in Hong Kong (154.44.31[.]4, AS979, NetLab Global) and hosted an attack egress and staging node in the U.S.United StatesU.S. defense policy
Industries
Civil societyHuman rights and civil society: Four organizations including a Canadian human-rights body, a U.S.Defensedefense policyEducationteachers' union, a science-policy advocacy group and an education nonprofit.Human rightsHuman rights and civil society: Four organizations including a Canadian human-rights body, a U.S.semiconductor manufacturingAdvanced semiconductor manufacturing
