Bloomberg and The Neuron Daily reported on campaigns in South Korea and Japan using open-source AI pentesting agents
SharkNinja Confirms Breach as AI Hacking Tool Campaigns Span US, Asia
Exposed control panels give defenders direct visibility into how AI agents are being operationalized for financially motivated intrusions at scale, with confirmed corporate victims and valid credentials still in circulation. The campaigns show two distinct open-source AI tools repurposed from legitimate penetration-testing use for criminal intrusions across multiple countries.
The full picture
SharkNinja confirmed a breach of a standalone product support site after NorForge uncovered more than 20 openly accessible control panels for Cairn, an open-source AI platform being used to conduct automated intrusions against organizations in the US, Japan, and Australia. NorForge identified two attacker clusters using the exposed panels: a Chinese-language group with 14 panels and over 1,100 attack projects targeting e-commerce and payment systems, and an English-language group with 10 panels focused on credit cards and cryptocurrency. The most active attacker breached more than 100 sites, with successful intrusions taking roughly 23 hours and costing approximately $5 in model fees. The dashboards were left open with no password, exposing hundreds of credentials of which 11% were verified as still valid; worker names suggest agents ran on DeepSeek's V4.1 Flash model. Cairn's creator stated the platform was never intended as a hacking tool and that misuse is a human decision. NorForge is working with Reuters to notify affected organizations and pursue removal of the malicious panels.
In a parallel campaign, CrowdStrike attributed a series of data-exfiltration attacks against South Korean financial institutions to an operator using ARTEX, a Chinese-developed open-source AI penetration-testing agent, active from late September to early October 2026. CrowdStrike assessed with moderate confidence that the operator was Chinese-speaking and financially motivated, based on Chinese-language prompts and use of the China-developed tool. Specific targets included a loan-progress inquiry service used by financial brokers and an employee mobile work-support system at separate banks. CrowdStrike stated that the AI models did not independently conduct the breaches and that a human operator used AI-enabled tooling alongside attacker-controlled infrastructure and other offensive techniques. Separately, Ierae examined Japanese attack infrastructure that targeted approximately 20 companies, nearly all in Japan and concentrated in a single industry.
How it developed
Unite.AI reported NorForge's full findings on 20+ exposed Cairn panels
CrowdStrike linked South Korean bank breaches to operator using ARTEX; campaign assessed active from late September to early October 2026
Sources
1 more source
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free