Post

Cairn and CL-STA-1238: autonomous AI attacks on 19 organizations

Unit 42 reports that China-linked CL-STA-1238 combined the open-source Cairn framework with Zhipu GLM and DeepSeek to run 25 autonomous operations. Unauthorized access was confirmed at 10 organizations.

In research published November 12, 2026, Unit 42 described a concrete case of autonomous AI agents used in targeted attacks. CL-STA-1238, assessed with moderate confidence as China-linked, combined the open-source penetration-testing framework Cairn with Zhipu GLM and DeepSeek models.

The result: at least 25 operations against 19 organizations between August and September 2026. Unauthorized access or data exfiltration was confirmed in 10 of them. Targets included semiconductor manufacturing, U.S. defense policy, and civil society organizations.

This post explains how the campaign worked, what Cairn does, and what it means for defenders.

Quick summary

Field Value
Actor CL-STA-1238 (Unit 42: moderate confidence China-linked)
Period August–September 2026
Tooling Cairn + Zhipu GLM/DeepSeek
Scale 25 ops, 19 orgs targeted, 10 confirmed compromises

What Cairn brings

Cairn is an offensive security automation framework: it can help chain recon, exploitation steps, and post-exploitation actions. Paired with capable LLMs, parts of an intrusion workflow can be planned and executed with less human-in-the-loop time—especially for repeatable tasks.

Defender takeaways

  • Assume faster iteration: AI-assisted operators can try more paths quickly; detection must focus on behaviors (lateral movement, credential abuse, anomalous data flows), not just known signatures.
  • Harden identity and egress: Least privilege, MFA, token hygiene, and monitored outbound channels reduce blast radius.
  • Threat intel ≠ silver bullet: Track actor TTPs, but prioritize controls that break common kill chains.

Autonomy in attacks is a spectrum; this case is a signal that “agentic” tooling is moving from demos into real operations.