CrowdStrike finds agentic AI tool ARTEX used against South Korean finance firms
CrowdStrike identifies ARTEX and multiple AI models in attacks targeting South Korean financial institutions, using exposed attacker records while leaving attribution unresolved.
The investigation links ARTEX and several language models to data-theft activity targeting financial organisations. Exposed attacker records reveal the tooling, but the operator’s identity remains unconfirmed.
CrowdStrike Intelligence reported on 7 October 2026 that an unidentified attacker used the agentic AI tool ARTEX alongside large language models in a campaign targeting South Korean financial organisations. The activity ran from late September into early October and involved data exfiltration, according to the company’s investigation. The full number of affected institutions and the attacker’s identity remain unconfirmed.
In its published investigation, CrowdStrike described finding attacker-controlled infrastructure containing ARTEX configuration files, Claude Code session histories and memory files. These records provided evidence of how AI-assisted tools were used alongside conventional offensive infrastructure.
Exposed attacker files reveal the AI toolchain
CrowdStrike traced the activity through two servers. One hosted an ARTEX instance, while a second, based in Hong Kong, contained operational records and configuration files. The exposed material included Chinese-language instructions directing AI-assisted penetration-testing activity.
ARTEX was released as an open-source agentic penetration-testing tool developed in China. CrowdStrike found that its primary model backend was DeepSeek v4.1-flash. Other Claude Code sessions involved GLM-5.3 and Grok 4.6, demonstrating the use of multiple AI services rather than one model operating independently.
The recovered session histories also revealed proxy infrastructure and interactions associated with the campaign. CrowdStrike published technical indicators to support further investigation by security teams.
Financial institutions were targeted through different systems
According to industry reporting cited by CrowdStrike, one affected bank experienced a compromise involving a loan-progress inquiry service used by financial brokers. Another incident involved an employee mobile work-support system.
The security company associated the activity with reported breaches through overlapping infrastructure and recovered operational records. However, it did not establish a definitive count of compromised institutions.
South Korean financial authorities and police were also investigating the incidents, as reported by Yonhap News Agency. Those investigations should not be treated as independent confirmation that ARTEX was responsible for every reported bank breach.
Attribution remains unresolved
CrowdStrike assessed with moderate confidence that the operator was likely Chinese-speaking and financially motivated. Its assessment drew partly on Chinese-language prompts and requests for information about selling stolen Korean data.
The investigation also uncovered possible personal details in an AI-generated résumé request. CrowdStrike cautioned that the information could not be conclusively linked to the attacker.
The findings therefore do not establish a named perpetrator, verified nationality or state sponsorship. The confirmed scope of the investigation is narrower: attacker-controlled infrastructure and exposed records linking ARTEX and multiple language models to a campaign against South Korean financial organisations.