CL-STA-1238: Chinese-speaking actors use AI pentest framework Cairn to run 25 intrusions against 19+ organizations, breaching 10
Matéria consolidada. Reúne 8 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 5 fontes trouxeram em 17/11/2026 12:04; os links antigos trazem para cá.
Fontes: securityonline.info ↗ darkreading.com ↗ gbhackers.com ↗ helpnetsecurity.com ↗ unit42.paloaltonetworks.com ↗
Between August and September 2026, Unit 42 (Palo Alto Networks) tracked a China-nexus intelligence-collection campaign, CL-STA-1238, in which attackers used the open-source AI penetration-testing framework Cairn to run intrusions autonomously. Unit 42 assesses the China-nexus attribution with moderate confidence (securityonline.info). The operation comprised at least 25 attempts against at least 19 organizations in semiconductor and advanced manufacturing supply chains, defense and defense policy, and human rights and civil society (unit42.paloaltonetworks.com, helpnetsecurity.com, gbhackers.com). Ten organizations suffered unauthorized access or data exfiltration (gbhackers.com, securityonline.info). Impact included more than one million personal records leaked, including member records at a U.S. teachers' union (helpnetsecurity.com, securityonline.info), military and government identities retrieved, and persistent administrator access created (unit42.paloaltonetworks.com). Exposed data included sensitive employee, student and government data (gbhackers.com). Unit 42 assesses the activity was likely carried out by a single operator or a small team (helpnetsecurity.com). It presents the campaign as a new paradigm of AI-driven autonomous attacks in which automation increased volume and coverage, not sophistication. A later report (gbhackers.com, 2026-11-13) shows how brief operator instructions were turned into sustained intrusion attempts. According to securityonline.info (2026-11-17), operators supplied only a root domain and a one-line objective in Chinese, and the agents planned and executed the intrusions. Dark Reading (darkreading.com, 2026-11-16) also reported that China-nexus actors breached 10 organizations in manufacturing, defense and civil/human rights sectors using Cairn. That item is only a headline-level summary and adds no new details; it names no actor, initial access vector, CVEs, IOCs or victims.
Detalhes técnicos
- Tooling: The actors used a customized fork of Cairn with an added skill pack (unit42.paloaltonetworks.com), hosted on a Hong Kong server (helpnetsecurity.com), which acted as the controller. Cairn autonomously planned and executed multi-stage attacks. It ran autonomous reasoning cycles through its blackboard system, and the operators injected operational hints (gbhackers.com). The blackboard is a shared architecture: worker agents alternate between Reason and Explore phases, and confirmed findings are stored as immutable facts (securityonline.info).
- Skill pack: Each session loaded the "AIScan Runner Skills" pack and generated an AUTHORIZATION.md file (securityonline.info).
- AI models: Zhipu GLM and DeepSeek models, reported as GLM 5.2 and DeepSeek V4 variants (DeepSeek V4 Flash/Pro per gbhackers.com), were configured with minimal prompts (helpnetsecurity.com). Per securityonline.info, 13 operations ran on Zhipu GLM 5.2, 8 on DeepSeek V4 Flash and 4 on DeepSeek V4 Pro; operators moved to Pro when a target resisted.
- Defense bypass: The attackers used a forged authorization pretext to get past the models' safeguards and defenses (gbhackers.com). Specifically, a Chinese system prompt claimed an authorized assessment or CTF to bypass model guardrails (securityonline.info).
- Kill chain: Operations covered reconnaissance, exploitation, credential theft, privilege escalation and persistence (gbhackers.com).
- Techniques: The attackers used credential theft, weak-password spraying, credential brute forcing, authentication bypasses, exposed cloud APIs, default credentials, and insecure cloud storage and endpoints. These are classic identity and configuration weaknesses, not novel exploits. Per securityonline.info, the agents brute-forced a partner-integration credential, sprayed weak passwords and used VPN gateway addresses with shared group passwords. They created persistent admin accounts and an API token. They exploited an unauthenticated account-creation bypass in a CMS, an unauthenticated cloud-storage write flaw and an open registration endpoint. No CVEs are named in the sources.
- Duration: Individual operations ranged from 20 minutes to more than 50 hours (helpnetsecurity.com, gbhackers.com).
- Attribution basis: Simplified Chinese prompts, exclusive use of Chinese AI models, and target selection aligned with Beijing's priorities (securityonline.info).
- Infrastructure and indicators: Cairn server in Hong Kong: 154.44.31.4. U.S. staging node: 82.29.129.54 (helpnetsecurity.com, securityonline.info).
Remediação
- Strengthen identity hygiene: enforce MFA, eliminate default and weak credentials, and defend against password spraying and brute forcing (securityonline.info).
- Reduce the attack surface by locking down exposed cloud APIs, storage and endpoints, including restricting unauthenticated cloud-storage writes and open registration endpoints (securityonline.info).
- Patch known vulnerabilities promptly (unit42.paloaltonetworks.com).
- Audit production systems for unauthorized admin accounts and API tokens (securityonline.info).
- Hunt for and block the two IPs listed above.
- Palo Alto Networks says its security products protect customers against this activity (unit42.paloaltonetworks.com). It specifically recommends Advanced URL Filtering, Advanced IP Defense and Cortex XDR (gbhackers.com).
Linha do tempo
-
CL-STA-1238 uses Cairn framework with Zhipu GLM and DeepSeek models for autonomous intrusions, breaching 10 of 19 targets
ler a análise desta fonte ocultar
Between August and September 2026, CL-STA-1238, a Chinese-speaking cluster that Unit 42 assesses with moderate confidence as China-nexus intelligence collection, ran 25 autonomous operations against 19 organizations using a customized fork of the open-source Cairn framework. Operators supplied only a root domain and a one-line objective in Chinese, and the agents planned and executed the intrusions. Ten targets were breached, among them semiconductor supply-chain firms, defense policy groups, and a U.S. teachers' union with more than one million member records exposed. The report comes from the article alone: web research was not performed, and no vendor or CVE data is available.
Technical Details
- Cairn uses a shared "blackboard" architecture. Worker agents alternate between Reason and Explore phases, and confirmed findings are stored as immutable facts.
- The operators ran 13 operations on Zhipu GLM 5.2, 8 on DeepSeek V4 Flash and 4 on DeepSeek V4 Pro. They moved to Pro when a target resisted.
- Each session loaded the "AIScan Runner Skills" pack and generated an AUTHORIZATION.md file. A Chinese system prompt claimed an authorized assessment or CTF in order to bypass model guardrails.
- The agents brute-forced a partner-integration credential, sprayed weak passwords and used VPN gateway addresses with shared group passwords. They created persistent admin accounts and an API token. They exploited an unauthenticated account-creation bypass in a CMS, an unauthenticated cloud-storage write flaw and an open registration endpoint.
- Infrastructure: a Hong Kong server (154.44.31.4) and a U.S. staging node (82.29.129.54).
- Attribution rests on Simplified Chinese prompts, exclusive use of Chinese AI models, and target selection aligned with Beijing's priorities.
Remediation
- Enforce MFA and remove default and weak passwords.
- Restrict unauthenticated cloud-storage writes and lock down open registration endpoints.
- Patch known flaws.
- Block 154.44.31.4 and 82.29.129.54.
- Audit production systems for unauthorized admin accounts and API tokens.
-
China-nexus actors breach 10 organizations in AI-driven autonomous attacks on manufacturing, defense and civil society
ler a análise desta fonte ocultar
A report by Dark Reading says China-nexus actors breached 10 organizations in attacks driven by AI. The attackers again used a tool called Cairn to run the attacks autonomously. The targets were advanced manufacturing and defense organizations, plus civil and human rights organizations. The source gives only a headline and a one-line description. It does not name the threat actor, the initial access vector, the CVEs, the IOCs or the victims. No research was done beyond the article, so the details above are unverified and the operational impact is unclear.
-
Chinese-speaking hackers leverage Cairn AI framework for autonomous intrusions in semiconductor and defense sectors
ler a análise desta fonte ocultar
In August and September 2026, Chinese-speaking attackers used the open-source Cairn AI penetration-testing framework combined with Chinese-language AI models such as Zhipu GLM 5.2 and DeepSeek V4 Flash/Pro to conduct at least 25 autonomous operations targeting semiconductor manufacturers, defense-policy organizations, and civil-society institutions. The campaign showed how agentic AI can translate brief operator instructions into sustained intrusion attempts lasting from 20 minutes to over 50 hours, with confirmed unauthorized access and data exfiltration at ten organizations. Attackers leveraged autonomous reasoning cycles via Cairn's blackboard system, injected operational hints, and used a forged authorization pretext to bypass defenses. Infrastructure included a Hong Kong controller and a U.S.-based staging server. Operations involved reconnaissance, exploitation, credential theft, privilege escalation, and persistence, exposing sensitive employee, student, and government data. Palo Alto Networks recommends using its Advanced URL Filtering, Advanced IP Defense, and Cortex XDR solutions to mitigate such threats.
-
Chinese-speaking actors use autonomous AI framework Cairn for 25 cyber operations against 19 organizations, breaching 10
ler a análise desta fonte ocultar
Palo Alto Networks’ Unit 42 discovered a campaign in August-September 2026 where Chinese-speaking attackers used Cairn, an open-source AI penetration testing framework, customized and hosted on a Hong Kong server, to autonomously conduct 25 intrusion attempts against at least 19 organizations. Of these, 10 suffered unauthorized access or data exfiltration. Target sectors included semiconductor and advanced manufacturing supply chains, defense and defense policy, and human rights and civil society. The attackers leveraged AI models Zhipu GLM 5.2 and DeepSeek V4 variants configured with minimal prompts, combined with classic weaknesses such as authentication bypasses, exposed cloud APIs, and default credentials. Individual attack durations ranged from 20 minutes to over 50 hours, demonstrating how automation increased attack volume and coverage rather than complexity. Credential theft, password spraying, brute forcing, and exploitation of insecure cloud storage and endpoints enabled extensive data access, including over a million records at a U.S. teachers’ union. Indicators include a Hong Kong Cairn server IP 154.44.31.4 and a U.S. staging node 82.29.129.54. Unit 42 assesses these operations as China-nexus intelligence collection under tracking code CL-STA-1238, carried out likely by a single operator or small team.
-
Chinese-speaking attackers use autonomous AI framework Cairn for espionage against semiconductor, defense, and civil society sectors
ler a análise desta fonte ocultar
Between August and September 2026, attackers linked to China-nexus intelligence campaign CL-STA-1238 used the open-source AI penetration testing framework Cairn, paired with Chinese AI models Zhipu GLM and DeepSeek, to autonomously conduct 25 operations targeting at least 19 organizations across semiconductor manufacturing, defense, and civil society sectors. Cairn autonomously planned and executed complex multi-stage attacks, achieving unauthorized access and data retrieval at 10 organizations, including leaking over one million personal records, retrieving military and government identities, and creating persistent administrator access. The attackers used a customized fork of Cairn with an added skill pack, employed weak-password spraying, credential brute forcing, and exploited identity and configuration weaknesses. This campaign illustrates a new paradigm in AI-driven autonomous attacks, emphasizing the importance of strong identity hygiene, attack surface reduction, and patching against known vulnerabilities. Palo Alto Networks offers multiple security solutions to protect customers from such threats.