HighBreach / campaignMonitoring

Cairn AI framework abused in attacks on 10 organizations (CL-STA-1238)

Chinese-speaking actors used the open-source Cairn AI framework against at least 19 organizations; check logs for two IPs and fix basic identity gaps.

Published
Updated
Exploitation
Yes, Unit 42 findings relayed by three outlets report confirmed unauthorized access or data retrieval at 10 organizations

What happened

Palo Alto Networks’ Unit 42 tracks the activity as CL-STA-1238 and assesses with moderate confidence that it is China-nexus intelligence collection, according to Cyber Security News and GBHackers. GBHackers adds that the attribution rests on Chinese-language operational artifacts, exclusive use of Chinese AI models and targeting that fits Chinese intelligence priorities, not definitive evidence naming a government sponsor. The original Unit 42 post was not among the sources supplied, so the details below come from the three outlets.

  • Scale: 25 operations against at least 19 organizations in August and September 2026, with unauthorized access or data retrieval at 10 of them, per Dark Reading. Dark Reading calls this the second documented attack in which China-linked actors used Cairn. The other two outlets do not say this.
  • How it worked: Operators supplied a target domain and an objective. Cairn then planned, ran tests and recorded findings, according to GBHackers. AI workers share a “blackboard” where confirmed findings are stored as facts and proposed actions as intents. Human operators could inject hints, such as VPN addresses and shared passwords obtained earlier.
  • Models: 13 operations used Zhipu GLM 5.2, eight used DeepSeek V4 Flash and four used V4 Pro (GBHackers, Cyber Security News). Operations lasted from 20 minutes to more than 50 hours.
  • Guardrails: Cairn’s stock prompts already tell workers never to refuse tasks. The attackers added Chinese-language text claiming authorized security testing and generated an AUTHORIZATION.md file per task. Neither model refused, per Cyber Security News and GBHackers. GBHackers also reports that the deployment loaded a scanning pack called “AIScan Runner Skills” that is not in upstream Cairn.
  • Targets: Cyber Security News reports seven targets in the semiconductor and advanced-manufacturing supply chain, three defense-related and four human rights or civil society groups. Dark Reading gives seven and three for the first two groups and describes the rest only as human rights and civil society organizations. The sources do not name the victims beyond type, and the counts are not reconciled with the 19 total.
  • Reported outcomes:
    • An industry standards body: password spraying yielded dozens of valid credentials, and a live standards-ballot document was overwritten (GBHackers, Cyber Security News).
    • A semiconductor company: a brute-forced production credential on a partner-integration application, unauthorized administrator accounts and a persistent API token (Dark Reading, Cyber Security News).
    • A U.S. teachers’ union: an account-creation bypass on a content management system exposed more than one million member records, about 31,000 CRM contacts and staff password hashes.
    • A defense-policy program: an open registration endpoint exposed roughly 500 user records, including military and government email identities.
    • Other cases: employee data, student data and about 150,000 internal industrial-automation documents.
  • Not novel weaknesses: Unit 42, as quoted by Dark Reading, lists authentication bypasses, insecure direct object references, mass-assignment privilege escalation, weak and default credentials, exposed cloud APIs and dated public CVEs. No specific CVE identifiers are given in the sources. AI raised speed and reach, not the complexity of the attacks.

Am I affected?

The sources do not describe a flaw in Cairn itself, and there is no patch to apply. The risk is to organizations whose internet-facing systems carry the familiar weaknesses listed above. Whether your organization was targeted is not confirmed by any source unless you find the indicators in your own logs.

Indicators reported by GBHackers and Cyber Security News, defanged:

IP addressRoleLocation and hosting
154.44.31[.]4Cairn hosting serverHong Kong, AS979, NetLab Global
82.29.129[.]54Staging server and attack egressUnited States, AS142036, Hosteons Pte. Ltd.

Cyber Security News also reports that Unit 42 saw Cairn generate hundreds of thousands of high-frequency requests from one origin.

What to do now

  1. Search web, VPN, identity and cloud-storage logs for the two IP addresses above. Re-fang them only inside controlled tools such as a SIEM or MISP.
  2. Look for unusually noisy traffic from a single IP address, as Cyber Security News advises.
  3. Use strong, unique passwords with phishing-resistant MFA, and check for weak or default credentials, particularly on partner-integration apps and registration endpoints (Cyber Security News).
  4. Reduce internet-facing systems and patch known CVEs quickly (Cyber Security News). Unit 42 says identity hygiene, attack-surface reduction and patching remain effective, per Dark Reading.
  5. Consider continuous attack-surface management and deception “tripwires” to catch reconnaissance and lateral movement. These are suggestions from John Strand of Black Hills Information Security, quoted by Dark Reading, not Unit 42 findings.

Sources