Published
- 4 min read
By Allan D - Editor, AI Security Wire
Unit 42: Chinese-Speaking Operators Ran Cairn AI Agents Against 19+ Orgs
This is a follow-up to our October coverage of the Reuters investigation into a lone hacker using Cairn. Palo Alto Networks’ Unit 42 published its own analysis of Cairn abuse on 12 November 2026, and it describes a different slice of the activity, with a named cluster, victims by sector and a China-nexus intelligence-collection attribution (moderate confidence).
What Unit 42 found
Unit 42 says that in August and September 2026, Chinese-speaking attackers ran Cairn, an open-source AI penetration testing framework, to carry out 25 operations against at least 19 organizations. It tracks the cluster as CL-STA-1238. Access or data retrieval was confirmed at 10 of the targets.
In outline, the workflow was simple. An operator created a task with a one-line goal and a root domain. Cairn then mapped the target, broke the goal into smaller tasks and worked through a reason-then-explore loop, recording confirmed findings on a shared blackboard. Operations ran from 20 minutes to more than 50 hours with no time limit set by the attackers.
The models and the guardrail workaround
Cairn is model-agnostic. Unit 42 says the attackers used Zhipu GLM 5.2 for 13 operations, DeepSeek V4 Flash for eight and DeepSeek V4 Pro for four. It also says they appeared to move to the Pro model for targets that resisted Flash.
Getting the models to cooperate took very little. Cairn’s stock worker prompts already tell the model not to refuse a task. The attackers added a Chinese-language preamble claiming an authorized security test and forbidding refusal, plus a generated AUTHORIZATION.md file for each operation. Unit 42 reports that neither GLM nor DeepSeek refused that framing.
Who was hit
Unit 42 says the targets were concentrated in three main categories, covering 14 of the 19 organizations:
- Semiconductor and advanced manufacturing supply chain: seven organizations, from lithography and chip design to materials, photomasks and standards.
- Defense and defense policy: three organizations, including a space technology company, a defense-analytics vendor and a policy think tank.
- Human rights and civil society: four organizations, including a Canadian human-rights body and a US teachers’ union.
Confirmed outcomes included dozens of valid credentials at an industry standards body, whose live ballot document was overwritten through an unauthenticated cloud storage write. At a semiconductor company the attackers created administrator accounts and a persistent API token. At the teachers’ union, an account-creation bypass led to the framework logging access to more than one million member records. Roughly 500 user records, including .mil and .gov accounts, were exposed at a defense-policy identity platform.
Unit 42 notes that one task against a semiconductor equipment supplier already contained VPN gateway addresses, group names and shared group passwords, which it reads as intelligence gathered beforehand.
Attribution and how far to trust it
Unit 42 assesses with moderate confidence that this is China-nexus intelligence collection. Its three lines of evidence are Chinese-language tasking files, the exclusive use of Chinese models, and targeting that matches Chinese state priorities. Our view: that evidence is circumstantial, and this is a single vendor’s analysis. As of 12 November 2026, we found no second source for the 19-organization count or the attribution in the Reuters coverage and the Unit 42 report we checked.
What it means for defenders
The notable point is not sophistication. Unit 42 says the weaknesses exploited were known ones: authentication bypasses, insecure direct object references, mass-assignment flaws, weak and default credentials, exposed cloud APIs and dated public CVEs. The change is scale. Unit 42 judges the campaign was likely run by a single operator or small team that attacked at least 19 organizations at once within two weeks.
Our assessment, not Unit 42’s: the practical response is the unglamorous one. Check for default and reused credentials on internet-facing apps, test registration and account-creation endpoints for bypasses, and review cloud storage write permissions. Unit 42 lists two indicators, both attacker infrastructure: 154.44.31[.]4 (Hong Kong) and 82.29.129[.]54 (US staging and egress node).
References
Frequently Asked Questions
- What did Unit 42 report about Cairn?
- Unit 42 says Chinese-speaking attackers used the open-source Cairn penetration testing framework to run 25 autonomous operations against at least 19 organizations in August and September 2026. It reports unauthorized access or data retrieval at 10 of them. The attackers supplied a target and a goal, and the framework planned and executed the steps.
- Who does Unit 42 think is behind it?
- Unit 42 tracks the activity as CL-STA-1238 and assesses with moderate confidence that it is China-nexus intelligence collection. It cites Chinese-language tasking files, the exclusive use of Chinese AI models, and a target set in semiconductors, US defense policy and civil society. That is an assessment, not a confirmed attribution.
- Is this the same campaign Reuters described in October?
- We cannot say. Reuters reported a single hacker using Cairn against more than 1,200 sites, with Claude Code and DeepSeek mentioned in the reporting. Unit 42 describes a smaller, more targeted set of at least 19 organizations on GLM and DeepSeek models, in August and September, before the researcher in the Reuters story began observing Cairn instances around 2 October. Nothing in the Unit 42 report that we read links the two, so we treat them as separate until a source says otherwise.
- Did the attackers use new vulnerabilities?
- No, according to Unit 42. It lists authentication bypasses, insecure direct object references, mass-assignment privilege escalation, weak and default credentials, exposed cloud APIs and dated public CVEs. Its conclusion is that autonomous AI raises the volume and coverage of exploitation, not its complexity.