Skip to main content
All incident reports
Incident report

Shinhan, KB Kookmin and Hana banks breached in AI-agent attacks

South Korean authorities are investigating breaches at several banks, including Shinhan, KB Kookmin and Hana, that officials believe involved AI agents. CrowdStrike says the ARTEX AI penetration-testing suite and Claude agents were used, while local reports cited by The Record put exposed personal data at 68,000 people across at least seven institutions.

· First published

The short version

  • The breach. South Korean officials are investigating breaches at several banks, including Shinhan, KB Kookmin and Hana, and believe AI agents played a role in at least some attacks. The Record reports at least seven financial institutions were affected.
  • The tooling. CrowdStrike says the attacks used ARTEX AI, an agentic penetration-testing suite developed in China, and Claude agents. Researchers found open directories containing Claude Code session histories, ARTEX configuration files and Claude memory files.
  • Data exposed. Local reports cited by The Record say personal data of at least 68,000 people was exposed, including borrowing history, incomes, phone numbers and names. Shinhan reportedly said about 25,000 customers were affected; BleepingComputer says credit card information was also exposed.
  • Response. South Korea's president pledged staff and money, the National Office of Investigation formed a 28-person team, and the government held an emergency meeting. ARTEX's developer has made the project closed-source and stopped updates, but derivatives remain available.

Timeline

DateEvent
30 SeptemberThe bank breaches first came to light, according to The Record.
6 OctoberPresident Lee Jae Myung says there are signs AI agents were used in at least some attacks.
6 OctoberThe National Office of Investigation creates a 28-person team to investigate.
10 OctoberBleepingComputer reports CrowdStrike's finding that ARTEX AI and Claude agents were used.

How the attacks were carried out

The Record reports that South Korean financial authorities found 33 IP addresses used in the hacks, linked to at least 12 countries including Japan, the US, Thailand, Vietnam and Hong Kong. Officials believe the Chinese cybersecurity tool ARTEX AI was used against the banks' systems. President Lee said it has become possible to use AI to hack with ease without specialised skills.

CrowdStrike says the activity involved ARTEX AI, which it describes as an agentic penetration-testing suite developed in China, and Claude agents. Its researchers found open directories holding Claude Code session histories, ARTEX configuration files and Claude memory files. They say the ARTEX instance used DeepSeek as its main large language model (LLM) backend, supplemented by other models for additional Claude Code sessions.

Who is affected

The Record, citing local reports, says at least 68,000 people had personal data exposed across at least seven financial institutions. The data included customers' borrowing history, incomes, phone numbers and names. Shinhan Bank reportedly said about 25,000 customers were affected.

BleepingComputer reports that the banks include Shinhan, KB Kookmin and Hana, and that personal data and credit card information were exposed, with system outages in some cases. It does not give a separate total for affected customers or say how much card data was taken.

Response and attribution

South Korea's government held an emergency meeting and called for immediate security measures for critical IT systems, according to BleepingComputer. President Lee pledged significant staff and money, and the National Office of Investigation set up a team of 28 investigators, The Record reports.

CrowdStrike says the attacker may be a 26-year-old Chinese person who studied at the South China University of Technology and lives in Maoming, Guangdong. Its researchers also found a 2007 date of birth in the same records and say the details are not reliable enough to confirm identity. They say the records showed no specific plan to monetise the stolen data.

After confirming ARTEX had been used in real-world attacks, its developer made the project closed-source and stopped updates. The current code has been used for English- and Korean-language derivatives, so it remains available in that form. The Record also mentions separate AI-agent incidents in Australia and at Hugging Face but does not link them to the South Korean breaches.

What is not known yet

  • The identity of the attacker or attackers, and whether the person described in the leaked records is responsible.
  • How many customers and institutions were ultimately affected, and whether the 68,000 and 25,000 figures overlap.
  • Whether credit card data was taken and, if so, how much; BleepingComputer reports exposure but gives no volume.
  • How the attackers gained initial access to the banks' systems; the sources describe the AI tooling but not the entry point.
  • Whether any stolen data has been sold or otherwise used; CrowdStrike says the records showed no specific monetisation plan.

What to do

If you operate critical financial or IT systems

  1. Check internet-facing systems and exposed storage for ARTEX configuration files, Claude Code session histories and Claude memory files, the artefacts CrowdStrike found in this campaign.
  2. Review logs for access to LLM API proxies or resellers such as xcai[.]pro and for unusual AI-agent activity.
  3. Carry out the immediate security measures South Korea's government called for on critical IT systems, starting with access to customer data.

If you use ARTEX AI or a derivative

  1. Remove or isolate ARTEX AI installations and any English- or Korean-language derivatives; the original project is no longer maintained and derivatives remain available.

For everyone else

  1. Treat AI agents and automated penetration-testing tools as privileged users and limit what customer data they can reach.
  2. Inventory AI agents, log their sessions and review them for data access and for attempts to find data-sales channels.
  3. Restrict and monitor outbound access to external LLM APIs and resellers.

Update log

  • First published.

Sources

New incident reports by email

One email a week with the reports published since the last one.

Want us to look at your domain?

Give us one apex domain and we'll send back a written snapshot of what is already visible from outside. No call required.