Cairn AI Framework Attacks: Unit 42 Says Attackers Paired DeepSeek and Zhipu GLM With an Open-Source Pentest Tool, Reached 10 of 19 Organizations, and What to Check This Week
Unit 42 says Chinese-speaking attackers paired the open-source Cairn framework with DeepSeek and Zhipu GLM, targeting at least 19 organizations and confirming access or data retrieval at 10. What is confirmed and what to do this week.
Where Worldwide
On this page 8 sections
Security researchers at Palo Alto Networks' Unit 42 say Chinese-speaking attackers used an open-source AI penetration testing framework called Cairn, paired with the DeepSeek and Zhipu GLM AI models, against at least 19 organizations from August to September 2026. Unit 42's investigation found that Cairn confirmed unauthorized access or data retrieval at 10 of them, and that matters to any business with servers, websites or remote access on the internet, because AI is making it cheaper for attackers to try more targets, for longer.
This is not a flaw in one product that you can patch and forget. It is a change in how attacks are run. The defense is making sure the weaknesses these tools look for are already closed.
What happened in the Cairn AI attacks
Unit 42 published its research on November 12, 2026. According to the report:
- Unit 42 tracks the activity as CL-STA-1238.
- From August to September 2026, Chinese-speaking attackers used Cairn, an open-source AI penetration testing framework.
- The attackers weaponized Cairn by pairing it with two Chinese AI models, Zhipu GLM and DeepSeek.
- Cairn was used against at least 19 organizations across 25 operations.
- Unit 42's investigation shows Cairn confirmed unauthorized access or data retrieval at 10 of the targeted organizations.
Security teams normally use penetration testing tools, with permission, to find weaknesses before attackers do. In this campaign, a tool of that kind was used to get unauthorized access to organizations that were the targets.
Who is behind CL-STA-1238?
Unit 42 assesses with moderate confidence that CL-STA-1238 represents a China-nexus intelligence collection campaign. Treat that as the researchers' assessment. "Moderate confidence" means it is their best reading of the evidence, not proof. The report describes the people running the operations as Chinese-speaking attackers.
If you read coverage that goes further than that, check whether the claim comes from Unit 42 itself.
What is confirmed, what is claimed and what is not known
| Point | Status |
|---|---|
| Cairn used against at least 19 organizations in 25 operations | Reported by Unit 42 from its investigation |
| Unauthorized access or data retrieval confirmed at 10 organizations | Reported by Unit 42 from its investigation |
| Cairn paired with DeepSeek and Zhipu GLM | Reported by Unit 42 |
| Neither AI model refused the attackers' framing | Reported by Unit 42 |
| A China-nexus intelligence collection campaign | Unit 42 assessment, moderate confidence |
| Operations lasted from 20 minutes to more than 50 hours | Reported by BankInfoSecurity |
Not known from the sources we verified:
- The names of the targeted organizations
- Which industries or countries they are in
- Which systems or weaknesses Cairn used to get in
- What data was taken, and how much
- How many of the 25 operations used each AI model
- Whether the affected organizations have since cleaned up and secured their systems
We will not fill those gaps with guesses. If Unit 42 or the affected organizations publish more, the picture may change.
How DeepSeek and Zhipu GLM were used in the attacks
Unit 42 says the attackers used a framing to get the AI models to help. In Unit 42's words: "Neither Zhipu GLM nor DeepSeek refused this framing." The details we verified do not say exactly what wording the attackers used.
Nothing in the sources we verified says the companies behind DeepSeek or Zhipu GLM were involved. The attackers used the models as tools inside Cairn. The lesson for a business is that an AI model's safety rules are not a security control you can rely on. If someone wants to misuse a model, the rules may not stop them.
Why AI-automated attacks matter for a business network
Unit 42 puts it this way: "Autonomous AI raises the volume and coverage of exploitation, but not its complexity."
In plain terms, the AI did not invent new ways to break in. It let the attackers try more organizations, and check more of each organization's exposed systems, than they could by hand. BankInfoSecurity reports that individual operations ran from 20 minutes to more than 50 hours, with no time limit set by the attackers.
A person working by hand gets tired, gets bored and moves on. An automated run keeps going. For a company network or website, that changes how fast an old weakness gets found:
- A forgotten test or staging site that nobody updates
- A remote access login left open to the whole internet
- A server running software that was updated on paper but never actually patched
- An admin panel that only has a password, with no second step
None of these are new problems. What has changed is that someone can now check every one of them across many targets at once. If you are not sure what your business has on the internet, start with Your Website Is Live. Can Anyone Find It, and Is It Locked? and Who Is Patching the Software That Runs Your Business?.
What to do this week about AI-automated attacks
The sources we verified do not name a specific product, version or setting that Cairn exploited, so there is no single fix to install. These steps cut down what an automated tool can find:
| What to do | How to check |
|---|---|
| List every system reachable from the internet | Go through firewall port forwards, DNS records, hosting accounts and cloud consoles. Include old test and staging sites. |
| Patch internet-facing systems first | Compare each installed version with the vendor's current security release, and confirm the update is installed, not just downloaded. See Updated Is Not the Same as Patched. |
| Take down what you no longer need | Remove old sites, unused remote desktop access and admin pages that are open to everyone. |
| Require multi-factor authentication | Check VPN, email, hosting and website admin accounts. A password alone should not be enough. |
| Look for long, steady probing in your logs | Look for repeated requests from the same sources over hours or days in firewall, VPN and web server logs. |
| Name who responds | Decide who reviews security alerts and who to call if something looks wrong. See When Your Systems Go Down, Who Do You Call? |
Cairn AI framework questions people are asking
What is the Cairn AI framework?
Cairn is an open-source AI penetration testing framework. Unit 42 reports that attackers paired it with the DeepSeek and Zhipu GLM AI models and used it in attacks from August to September 2026.
Is Cairn malware?
Unit 42 describes Cairn as an open-source penetration testing framework, which is a kind of tool built for security testing. The problem in this case is that attackers used it against organizations without permission.
Did DeepSeek hack these companies?
No source we verified says that. Attackers used DeepSeek and Zhipu GLM inside Cairn, and Unit 42 says neither model refused the attackers' framing.
How many organizations were hacked with Cairn?
Unit 42 says Cairn was used against at least 19 organizations in 25 operations, and confirmed unauthorized access or data retrieval at 10 of them.
Who was behind the Cairn attacks?
Unit 42 tracks the activity as CL-STA-1238 and describes the attackers as Chinese-speaking. It assesses with moderate confidence that this is a China-nexus intelligence collection campaign.
Which companies were targeted?
The sources we verified do not name the organizations, their industries or their countries.
Is there a patch for the Cairn attacks?
There is no single patch. Cairn is a tool, not a vulnerability. The sources we verified do not list the weaknesses it used, so the best defense is to patch what you have on the internet and remove what you do not need.
Are AI cyberattacks more advanced than normal hacking?
According to Unit 42, not in this case. Its finding is that autonomous AI raises the volume and coverage of exploitation, but not its complexity. The attacks reach more targets and run longer, but they still rely on weaknesses that already exist.
The bigger lesson
AI did not create a new kind of attack here. It let one group test more organizations, more thoroughly and for longer. The defense that still works is the same one as before: know what you have on the internet, keep it patched, put a second factor on every login that matters, and have someone watching. What has changed is how quickly an automated tool will find whatever you missed.
Want your internet-facing systems checked and hardened? See security hardening or get a quote.
Get security alerts like this by email, with the free checklist, at /security-alerts.
Sources: Unit 42, Palo Alto Networks: Cairn framework research; BankInfoSecurity.
Have a question about this article?
Ask it here and we will answer you directly by email. Questions are private, never posted.
Security alerts by email
Hear about the next one first.
When attackers start using a flaw in software businesses run on, we email what it is, whether it touches you and what to do that week, with a round-up of our latest security writing. Free, one-click unsubscribe. You get the checklist straight away: 12 things to secure this week.
Want a result like this?
Security alerts, breaches and the technology a business runs on: what happened, who it affects and what to do.
Get a Tech QuoteKeep reading
Automated AI Cyberattack: Google Finds exploit_agent Turning Software Patches Into Live Attacks With No Human in the Loop, and What to Do This Week
Google says attackers used an AI framework called exploit_agent to find and exploit flaws in popular web softw...
SignDrop Windows Zero-Day: Public Exploit Claims SYSTEM Access Through CLFS, No Microsoft Patch or CVE Yet, and What to Do This Week
A researcher has published SignDrop, exploit code they say turns a normal Windows login into full SYSTEM contr...
TSMC Data Breach: 46,000 Employee Records Reportedly for Sale, TSMC Finds No Evidence Its Systems Were Hit, and What to Check at Your Own Vendors
Anonymous sellers reportedly offered 46,000 TSMC employee records for three Bitcoins. TSMC says it found no ev...