Palo Alto Networks’ Unit 42 says Chinese-speaking attackers used an open-source AI framework called Cairn to run 25 operations against at least 19 organizations between August and September 2026. All 25 ran on Chinese models, by our sum of the split Unit 42 gives: 13 on Zhipu GLM 5.2, 8 on DeepSeek V4 Flash and 4 on DeepSeek V4 Pro.
Cairn sits next to ARTEX, the framework behind the financial-sector campaign Microsoft described on 8 November. Axios reported on 18 November that open-source tools like these are free to download and are changing the economics of cybercrime.
Business Pill · Least Privilege
An evergreen Business Engineer explainer on least privilege and access control. It teaches the general idea only and does not discuss these attacks.
The key insight: All 25 Cairn operations Unit 42 attributes to one group ran on Chinese models, by our sum, and the FBI says Chinese models are the most prevalent in attacker-run ARTEX instances; the FBI lists 10 sectors hit by ARTEX.
What the FBI Says About ARTEX
The FBI’s FLASH alert of 13 November calls ARTEX an open-source, LLM-driven agentic platform that threat actors have weaponized to automate intrusions. It says malicious actors have used it since at least August 2026, with some intrusions progressing from initial targeting to privileged access in under three hours.
The alert lists the sectors hit: financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications and legal. That is 10 sectors, by our count.
According to the FBI, the original developer made the GitHub repository unavailable on October 8, 2026, after reports of abuse. The alert says the framework still spreads through forks, archived copies and a commercial version, ARTEX PRO, and that researchers identified over 1,600 publicly exposed ARTEX console instances as of early November. We reported Censys’s count of 1,465 exposed consoles on 6 November.
The FBI says Chinese models are the most prevalent in attacker-controlled ARTEX instances, with DeepSeek Flash family models dominating usage.

What Unit 42 Found in the Cairn Operations
Unit 42 says Cairn is model-agnostic and that the attackers ran their operations on Zhipu GLM 5.2 and DeepSeek V4 Flash and Pro. It writes: “Neither Zhipu GLM nor DeepSeek refused this framing”, referring to the attackers presenting the task as a client-approved engagement.
Axios adds that Google reported on Monday that hackers used Cairn during an intrusion into the U.S. aerospace sector.
What Amazon and CrowdStrike Saw
Amazon Threat Intelligence says decoy sensors it runs captured requests from ARTEX agents between Oct. 3 and Nov. 5, 2026. It says the most prominent campaign was a large-scale theft of credentials to access AI models, in which stolen keys paid for the LLM inference powering further credential hunting.
CrowdStrike says a group it tracks as VAPOR PANDA used ARTEX against more than 200 distinct organizations worldwide beginning on October 17, 2026. Axios puts that campaign at more than 200 organizations over 11 days.
What Defenders Told Axios
Sam Rubin of Palo Alto Networks’ Unit 42 told Axios: “We’re seeing more active attacks than ever right now”. Google’s John Hultquist told Axios that some incident response firms are turning away cases because they lack resources.
This week’s other AI-security reporting in our archive covers Reflection AI gating its Laser hacking agent, the patching rush behind last week’s outages, James Forshaw’s Windows UAC bypass and the Java4Shell flaw found with Codex. On Chinese models, see the proposal to make US companies disclose their use.
The Structural Read
As we read the reports together, the frameworks are free and the models are cheap; Amazon describes stolen AI keys paying for the next round of credential hunting.
As we read the FBI alert, removing the original repository did not remove the tool: the alert counts forks, archived copies, a commercial version and over 1,600 exposed consoles after the takedown.
Sam Rubin, SVP, Palo Alto Networks Unit 42, to Axios
“We’re seeing more active attacks than ever right now”
Three Implications
FOR SECURITY TEAMS The FBI says some ARTEX intrusions went from initial targeting to privileged access in under three hours.
FOR AI MODEL PROVIDERS Unit 42 says neither Zhipu GLM nor DeepSeek refused the attackers’ client-approved-engagement framing.
FOR CLOUD CUSTOMERS Amazon says the most prominent ARTEX campaign it observed stole credentials to access AI models.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Agentic Harness War.
The analysis puts it this way: “Foundation models become relatively commoditized as orchestration layers capture differentiation.”
As we read the Unit 42 and FBI reports, ARTEX and Cairn are such orchestration layers: the attackers swapped between Zhipu and DeepSeek models underneath the same framework. The analysis predates these reports and does not discuss them.
What Is Not Established
None of the reports gives a total number of victims across all ARTEX and Cairn activity, and Amazon says activity against other organizations could not be confirmed. The 25-of-25 figure covers only the Cairn operations Unit 42 attributes to one group in August and September; it does not describe all Cairn or ARTEX use. We did not contact the companies or the FBI; statements are as published or as reported by Axios.
The Bottom Line
Unit 42 counts 25 Cairn operations by one group, all on Zhipu or DeepSeek models by our sum, and the FBI says Chinese models are the most prevalent in attacker-run ARTEX instances. The FBI lists 10 sectors hit by ARTEX and over 1,600 exposed consoles as of early November.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Figures and quotations come from the FBI FLASH alert of 13 November 2026, the Unit 42 report of 12 November, Amazon Threat Intelligence’s post of 13 November and CrowdStrike’s post of 29 October, each read at source, and from Axios’s report of 18 November. The 25-of-25 share and the count of 10 sectors are our arithmetic from those sources. We did not contact the companies or the FBI.
Sources: Palo Alto Networks Unit 42: Cairn framework report, 12 Nov 2026 · FBI FLASH-20261113-001: Threat Overview of ARTEX, 13 Nov 2026 · Amazon Threat Intelligence, AWS Security Blog, 13 Nov 2026 · CrowdStrike: VAPOR PANDA Uses AI-Driven ARTEX, 29 Oct 2026 · Axios: The terrifying rise of AI hacking tools, 18 Nov 2026









