TAG-217 used AI tools (ARTEX, Claude Code, DeepSeek, Grok) in preparations linked to South Korean bank breaches affecting 68,000+
Matéria consolidada. Reúne 2 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 2 fontes trouxeram em 23/10/2026 00:06; os links antigos trazem para cá.
Fontes: recordedfuture.com ↗ gbhackers.com ↗
Recorded Future's Insikt Group tracks the actor behind the October 2026 intrusions into South Korean financial institutions as TAG-217, a suspected Chinese-speaking operator. It is not confirmed as state-backed (gbhackers.com). South Korea's Financial Services Commission (FSC) disclosed the attacks in early October (recordedfuture.com). Public reporting cited by both sources says personal data of at least 68,000 people was exposed across at least seven institutions, with no single weakness shared by the banks. Industry sources linked the attacks to ARTEX, an open-source Chinese-language AI penetration testing framework, and the actor also used Claude Code, DeepSeek, Grok and Zhipu AI. In a report dated October 16, Insikt Group said operational security mistakes (open directories and exposed files holding logs, scripts, Claude Code sessions, ARTEX settings and AI memory files) revealed that TAG-217 was testing infrastructure and building tools as early as August 2026. These tools went beyond public reporting and included cryptocurrency theft, Linux privilege escalation and mobile exploitation testing. The evidence does not show that AI acted autonomously, and it cannot measure how much AI sped up the attacks.
Detalhes técnicos
- Aug 17–22, 2026: TAG-217 tested remote-access infrastructure and scanned internet-facing systems. Logs show test connections between its own servers and no successful access to victims (recordedfuture.com, gbhackers.com).
- By Aug 22: It had assembled tools targeting cryptocurrency mining operations on the YiiMP platform. They tried default passwords, stole wallet configuration and key files, kept remote access and sent Telegram alerts when credentials were captured. Other tools targeted known Linux flaws for privilege escalation or extraction of protected password data.
- By Sep 5: A workspace existed for testing iPhone weaknesses and modifying Android system software. Logs show experiments but no proven mobile compromise.
- By Sep 26: The environment held AI tools and records covering reconnaissance, traffic proxying, payload delivery and examination of payment processes. It also had tools routing traffic through residential internet connections (residential proxies, used to hide the actor's origin), maintaining links between servers and copying AI working-memory files between TAG-217 servers.
- Victim systems and exposed data differed per institution. One breach exposed an outsourced developer's personal information and another involved theft of customer data, which is consistent with opportunistic testing at scale (recordedfuture.com).
- The report names no single malware family behind every intrusion, and finding a capability does not prove it was used against a bank (gbhackers.com).
- IoCs are available only to customers of the Recorded Future platform and are not listed in the article (recordedfuture.com).
Remediação
- Inventory all internet-accessible services, including employee and partner applications, and give them the same attention as customer-facing services. Verify users can access only authorized records, and prioritize fixes by exposure and data at risk.
- Add logging and monitoring to internet-facing services, and address unresolved vulnerabilities and weak credentials.
- Feed continuously updated threat intelligence into monitoring and response systems to enrich alerts and trigger retrospective searches.
- Correlate attacker infrastructure intelligence with unusual sign-ins, repeated requests for different customers' records, unexpected remote access and abnormal transfers. These signals still work when attackers rotate proxy addresses (gbhackers.com).
- Stop confirmed malicious sessions, and review uncertain findings before taking actions that could disrupt customers or shared services (gbhackers.com).
Linha do tempo
-
TAG-217 used AI-assisted ARTEX framework against South Korean banks, with crypto-theft, Linux and mobile tooling found as early as August 2026
ler a análise desta fonte ocultar
Recorded Future's Insikt Group tracks the actor behind the October 2026 AI-assisted intrusions into South Korean financial institutions as TAG-217. The group was testing infrastructure and assembling attack tools as early as August 2026, weeks before the campaign became public. South Korea's Financial Services Commission (FSC) disclosed the attacks in early October. Public reporting indicates the breaches exposed personal data of at least 68,000 people across at least seven institutions. Industry sources linked the attacks to ARTEX, a Chinese-language open-source AI penetration testing framework. Public analysis suggests a likely Chinese-speaking operator who also used Claude Code, DeepSeek, Grok and Zhipu AI. Insikt Group found operational security mistakes (open directories and exposed files) that revealed broader capabilities than public reporting, including cryptocurrency theft, Linux privilege escalation and mobile exploitation testing. The evidence does not establish how much AI accelerated the attacks.
Technical Details
- Aug 17–22, 2026: TAG-217 tested remote-access infrastructure and ran internet scanning. Logs show test connections between its own servers, with no successful access to victims.
- By Aug 22: It had assembled tools targeting cryptocurrency mining operations on the YiiMP platform. They could try default passwords, steal wallet configuration and key files, maintain remote access and send Telegram alerts when credentials were captured. Additional tools targeted known Linux weaknesses for privilege escalation or extraction of protected password data.
- By Sep 5: It had a workspace for testing iPhone weaknesses and modifying Android system software. Logs show experimentation but no demonstrated compromise.
- By Sep 26: Its environment held AI tools and records referencing reconnaissance, traffic proxying, payload delivery and examination of payment processes. It also had tools routing traffic through residential internet connections, maintaining links between servers and copying AI working-memory files between TAG-217 servers.
- The victims' systems and exposed data differed per institution. One breach exposed an outsourced developer's personal information and another involved theft of customer data. This is consistent with opportunistic testing at scale.
- IoCs are available only to customers of the Recorded Future platform and are not listed in the article.
Remediation
- Inventory all internet-accessible services, including employee and partner applications. Verify users can access only authorized records, and prioritize fixes by exposure and data at risk.
- Add logging and monitoring to internet-facing services, and address unresolved vulnerabilities and weak credentials.
- Feed continuously updated threat intelligence into monitoring and response systems to enrich alerts and trigger retrospective searches.
- Correlate attacker infrastructure intelligence with internal signals such as unusual sign-ins, repeated requests for different customers' records, unexpected remote access and abnormal data transfers.
- Preauthorize proportionate automated responses for well-understood threats, with review required when evidence is uncertain or the address belongs to a shared proxy service.
-
TAG-217 used Claude Code, DeepSeek, Grok and ARTEX in AI-assisted preparation linked to South Korean bank breaches
ler a análise desta fonte ocultar
Recorded Future's Insikt Group (report of October 16) tracks a suspected Chinese-speaking actor, TAG-217, linked to attacks on South Korean financial institutions. The actor used Claude Code, DeepSeek, Grok and Zhipu AI alongside ARTEX, an open-source Chinese-language AI penetration testing framework. Public reporting cited in the investigation says the breaches exposed personal data of at least 68,000 people across at least seven institutions, but there was no single shared weakness across the banks. Open directories on attacker infrastructure exposed logs, scripts, Claude Code sessions, ARTEX settings and AI memory files, which let researchers trace the preparations. The evidence does not show that AI acted autonomously, and it cannot measure how much AI sped up the attacks. The actor is not confirmed as state-backed.
Technical Details
- Aug 17-22: TAG-217 tested remote-access infrastructure and scanned internet-facing systems. The logs do not show successful access to victims.
- By Aug 22: the actor had assembled tools targeting YiiMP cryptocurrency mining operations. They tried default passwords, stole wallet configuration and key files, kept remote access, and sent Telegram alerts after capturing credentials. Other tools targeted known Linux flaws for privilege escalation or extraction of protected password data.
- By Sep 5: a workspace existed for testing iPhone weaknesses and modifying Android system software. The logs show experiments, not proven mobile compromises.
- By Sep 26: the environment held AI tools and records covering reconnaissance, proxy traffic, payload delivery and checks of payment processes. The operator also prepared ways to copy AI working-memory files between servers.
- The actor used residential proxies to hide its origin.
- The report names no single malware family behind every intrusion. Finding a capability does not prove it was used against a bank.
Remediation
- Identify internet-accessible services, review access permissions, and fix weaknesses according to the data and systems at risk.
- Give employee and partner applications the same attention as customer-facing services.
- Correlate infrastructure intelligence with unusual sign-ins, repeated requests for different customers' records, unexpected remote access and abnormal transfers. These signals still work when attackers rotate proxy addresses.
- Stop confirmed malicious sessions, and review uncertain findings before taking actions that could disrupt customers or shared services.