Unit 42 documents low-skill actor using jailbroken DeepSeek V4 to breach Indian university, stealing 5,200+ records; abliterated DeepSeek V4.1 Flash linked to South Korean bank attacks
ConcludedRan 6 daysin the edition since 24 Oct 20267 sourcesupdated 29 Oct 2026
Chinese AI agent website breach campaign
The gist
Latest: AI Agent Campaign Breached 100+ Firms at $50 Per Intrusion
The campaign shows autonomous AI agents conducting financially motivated intrusions at a cost and scale that changes the economics of cybercrime. Named companies have confirmed breaches or active investigations, and hundreds of exposed credentials were found to still be valid.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
A Chinese-speaking threat actor used the open-source Cairn AI agent framework, combining DeepSeek models with a Claude Code harness, to breach more than 100 companies including John Deere, SharkNinja, and Cummins, probing over 1,200 sites across the United States, Canada, Australia, and Japan. The campaign began September 25, 2026, with attack instructions issued in fewer than 100 Chinese characters per target. NorForge discovered the operation through more than 20 publicly exposed Cairn control panels, found via Shodan and Censys with no password protection, which revealed the operational details of ongoing campaigns. Successful intrusions averaged roughly 23 hours each, and the entire campaign cost approximately $5,000 in AI token spend, putting the per-breach cost at around $50. NorForge estimates about 10% of projects succeeded. Cairn operates as a state-based problem-solving engine in which agents post facts and planned actions to a shared blackboard, with human operators able to seed the process with hints. NorForge's investigation was prompted by a September 2026 Gambit Security report on criminal campaigns using AI harnesses including Hermes, Strix, and Cairn.
How it developed
The Neuron Daily briefly referenced NorForge's Cairn dashboard findings in a separate newsletter
First in the edition · 24 October 2026
Reuters reporting confirmed that the attacker in the Cairn panel campaign, which breached more than 100 companies beginning September 25, 2026, combined DeepSeek models with a Claude Code harness and issued attack instructions in fewer than 100 Chinese characters per target.
NorForge researcher Snorre Fagerland, now named on the record, said: "It's a new world now".
First citedZvi's AI Roundups
Unit 42 publishes report on Chinese-speaking threat actor using DeepSeek and Hermes Agent for data exfiltration, VoIP fraud, and WordPress exploitation
NorForge found more than 20 open Cairn control panels on Shodan and Censys, exposing a Chinese-speaking operator's use of autonomous AI agents to breach over 100 websites beginning September 25, 2026, across the United States, Canada, Australia, and Japan for roughly $5,000 total.
The dashboards contained hundreds of credentials, 11% of which NorForge verified as valid. SharkNinja confirmed its breach was limited to an isolated product-support site, and Cummins said it was investigating a potential attack on a third-party vendor's website.
NorForge findings published; The Verge, QZ, Unite.AI, and OpenDataScience reported on exposed Cairn panels and confirmed breaches at Deere, SharkNinja, and Cummins
CrowdStrike linked South Korean bank breaches to operator using ARTEX; campaign assessed active from late September to early October 2026
Sources
The Neuron
opendatascience.com
shattered.io
Related
- The ARTEX South Korean bank breach
- OpenAI agent breach of Australian Medicare
- Criminal use of jailbroken DeepSeek
- OpenAI agents' malicious RubyGems campaign
BackgroundAI security: attacks and defenses
Get the daily email
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.