Skip to content

Threat Actors

Amazon Finds China-linked Actor Using Cairn Against Universities

Amazon Threat Intelligence found an exposed Cairn server on September 10, 2026, and tied the activity to a Chinese-nexus threat actor.

https://breach-signal.com/threat-actors/amazon-cairn-china-actor-zero-day-universities/

Share

Key facts

  • Amazon Threat Intelligence discovered a publicly accessible server running Cairn, an open-source AI penetration-testing framework, on September 10, 2026.
  • A China-linked threat actor used Cairn to target universities in East Asia outside mainland China between September 8 and 10, 2026.
  • Cairn’s AI agents discovered and exploited a previously unknown zero-day vulnerability in MONAI Label, an open-source medical imaging AI application, within 11 minutes.
  • Amazon Threat Intelligence assessed with high confidence that the MONAI Label vulnerability was a zero-day, as no prior fixes, advisories, or mentions existed in vulnerability databases.

Amazon Threat Intelligence found an openly accessible server operating Cairn, an open-source AI penetration-testing framework, on September 10, 2026, revealing a China-linked campaign against universities in East Asia outside mainland China between September 8 and 10, 2026.

The AI agents behind Cairn found and exploited an unknown zero-day flaw in MONAI Label, an open-source medical imaging AI application, in just 11 minutes. Amazon Threat Intelligence said with high confidence that the flaw was a zero-day, noting that no prior fixes, advisories, or mentions of it appeared in vulnerability databases. The path traversal flaw allowed agents to write files to any location on the server and achieve remote code execution.

After establishing a proxy into a second university network running the same vulnerable software, the operator directed Cairn agents to target other machines with the zero-day. Amazon Threat Intelligence contacted the maintainers of MONAI Label, who fixed the bug and later released a security update on November 18, 2026.

Amazon attributed the campaign to a Chinese-nexus threat actor with moderate confidence, citing Simplified Chinese prompts, China Standard Time timestamps, and its choice of targets. The company shared two IP addresses linked to the attacks, 142[.]4.98.145 and 142[.]4.107.34, observed between September 8 and 10, 2026. The attackers had left the Cairn instance publicly accessible without authentication, exposing their operational history, prompts, and AI outputs.

The campaign also included failed efforts. Cairn agents attempted 148 actions over 16 hours to exploit the Log4Shell vulnerability and brute-force credentials on a Redis server without success. In a separate case, agents tested more than 30,000 credential combinations tied to 80 real usernames against university servers across more than six hours, and against a university GPU server for six hours, without gaining access.

Cairn works by coordinating AI agents through a shared blackboard that records findings and queues next actions as a fact-intent graph. In the Unit 42 campaign, operators gave Cairn a target and a high-level goal; the framework then built its own plan, ran reconnaissance, spotted vulnerabilities, tried exploitation, and logged its findings. Human operators can inject hints at any time to guide the operation.

The framework’s default worker prompts tell the model never to refuse a task. Attackers layered on an authorization pretext, creating an authorization.md file for each operation, and seeded each task with an organization’s root domain so the system could map the attack surface on its own. The tool requires only a starting point and objective in plain language.

Palo Alto Networks Unit 42 followed 25 autonomous AI-driven attacks that used Cairn against 19 organizations from August through September 2026. The actor, known as CL-STA-1238, broke into 10 of the 19 organizations, taking credentials, exposing more than 1 million records, and altering critical documents. Unit 42 called it the second documented attack in which China-linked actors automated their operations using Cairn.

Among the 19 targets, seven belonged to the semiconductor and advanced-manufacturing supply chain and three to US defense and defense policy; human rights and civil society groups were also targeted. Unit 42 pointed to an industry standards body struck by password spraying, a semiconductor company where a production credential was brute-forced, a US teacher’s union hit through an unauthenticated account-creation bypass, and a defense policy think tank targeted via an open back-end registration endpoint. Tyler Oliver, CEO and founder of Huntbase, said, “Either way, the password spraying looks like password spraying.”

The campaign also went after content management systems by exploiting identity and configuration weaknesses at a semiconductor-lithography manufacturer, an industrial-automation manufacturer, and an education nonprofit. Attackers grabbed military and government identities at other organizations and created persistent admin accounts in multiple product systems. The attackers obtained unauthenticated access to cloud storage, changed a live ballot for an industry standards vote, and signed in as arbitrary customers with read, write, and delete access to meeting, voting, and back-office functions.

The attackers operated a customized fork of Cairn from a Hong Kong server and hosted an attack egress and staging node in the US. Individual operations lasted from 20 minutes to more than 50 hours, and attackers ran simultaneous operations against multiple organizations. Every operation generated a different plan involving between three and seven stages, with no time limit imposed by the attacker.

The attacks paired Cairn with Chinese large language models Zhipu GLM 5.2 and DeepSeek V4 Flash/Pro, allowing the system to operate without human intervention. Palo Alto Networks Unit 42 said Cairn attacks depend on known weaknesses such as authentication bypasses, weak credentials, and exposed APIs, and said identity hygiene, attack-surface reduction, and software patching remain effective defenses.

Andy Piazza of Palo Alto Networks Unit 42 said Cairn automation allows threat actors to strike more targets faster and perform technical tasks without learning the underlying technologies. He said the monitored campaign can interact with 1,000 organizations at the same time, an approach that breaks perimeter risk models and attribution models.

Google researchers saw intrusions using Cairn aimed at U.S. aerospace sector organizations. Google’s Threat Intelligence Group and Mandiant reported in a third-quarter 2026 review that threat actors were moving beyond AI chatbots to multi-agent frameworks for scanning systems, identifying weaknesses, and developing exploits. GTIG saw Cairn and PentAGI used in real-world activity, while other groups tested CyberStrikeAI and HexStrike AI.

GTIG also saw China-linked actors using AI tools to study targets and review data taken during intrusions. The FBI released an advisory on November 13, 2026, warning about Artex, an AI tool that automates cyber intrusions from reconnaissance through data exfiltration. Microsoft tied Artex to Chinese state-backed actors in a supply chain attack on IT management software company ConnectWise, and Artex was linked to attacks on South Korean and Japanese financial institutions.

The framework began as a research tool for defenders but threat actors now weaponize it because it is simple to use and adaptable. Gambit Security recorded Cairn’s use by a financially motivated hacking group that targeted online retailers in September 2026. Eyal Sela of Gambit Security said Cairn matters more than Artex because it is a general-purpose tool without internal cybersecurity structuring, making it less predictable.

Amazon warned that AI agents can exploit vulnerabilities faster than security teams can respond, and the company urged automated defenses. CJ Moses said organizations should prioritize high-confidence detections that trigger automated responses, with humans overseeing outcomes instead of reviewing every individual action. The agents in the university campaign ran continuously for 16 hours without breaks, sleep, or working-hour constraints.

John Strand of Black Hills Information Security recommends continuous attack surface management to find and limit internet-exposed services, plus cyber deception tripwires to detect attackers during reconnaissance and lateral movement. Unit 42 said Cairn represents a shift in how attackers apply AI to targeted offensive operations. Google said it relies on threat intelligence, model safeguards, and platform enforcement to identify and disrupt misuse of its AI services.

AWS worked with a government partner to notify the affected organizations but declined to name the universities or their country. Amazon declined to say which AI models were used with Cairn in the attacks. MONAI has released a patch for the zero-day and is requesting a CVE identifier.

Was this helpful?

Join the discussion

Please read our Comment Policy before commenting. First-time comments are held for moderation. By posting a comment you agree to our Terms of Use and Privacy Policy, which explain how your name, email address and comment are used.

Your email address will not be published. Required fields are marked *