South Korea Says AI Was Used in Recent Bank Hacks That Exposed 68,000 Customer Records

South Korea has confirmed what security researchers have been warning about for months: artificial intelligence is now doing the hacking. The South Korean government said this week it is “highly likely” that an AI tool called ARTEX was used in a wave of data breaches that hit more than seven financial institutions between late September and early October 2026, exposing the personal information of roughly 68,000 customers.

US cybersecurity firm CrowdStrike went further in a report published October 7, tracing the campaign to a likely single attacker — possibly a 26-year-old based in China’s Guangdong province — who combined ARTEX, a recently released open-source AI penetration-testing tool, with Anthropic’s Claude Code assistant to breach multiple banks in a matter of days.

Key Takeaways

  • South Korea’s government says it is ‘highly likely’ an AI tool called ARTEX was used in data breaches at more than seven financial institutions.
  • CrowdStrike traced the campaign to a likely China-based individual who paired ARTEX with Anthropic’s Claude Code to hit at least nine banks.
  • Around 68,000 customers’ data was exposed, and ARTEX’s developer has now pulled the open-source tool citing misuse by ‘bad actors’.

What happened

At least nine South Korean financial institutions have disclosed or been reported as targets since late September. Shinhan Bank said information attached to loan applications for about 25,000 customers was leaked — including names, phone numbers, and annual income. KB Kookmin Bank disclosed a breach affecting around 119 customers, plus smaller exposures at Hana Bank. Yegaram Savings Bank reported data affecting approximately 40,000 people. Other confirmed targets include BNK Busan Bank, Welcome Savings Bank, Hyundai Capital, and PFCT. Woori Bank and NH NongHyup Bank said they detected similar intrusion attempts and blocked them before any data escaped.

South Korea’s Financial Security Institute traced the attack IPs and server logs and found traces of ARTEX — an AI-driven tool published on GitHub by a developer using the handle “Autumn-27,” designed to automate reconnaissance, vulnerability scanning, and attack-path planning. ARTEX was built as a defensive security-testing tool; it became an offensive weapon.

How one person hit so many banks

CrowdStrike’s investigation found ARTEX configuration files, Claude Code session histories, and Claude memory files on attacker-controlled infrastructure. The picture that emerged: ARTEX, backed primarily by DeepSeek’s v4.1-flash model (with GLM-5.3 and Grok 4.6 used in some sessions), handled the automated probing — discovering vulnerabilities and compromising specific services across multiple banks in parallel, without a human operator manually attacking each target.

Claude Code served as the assistant. The attacker reportedly asked it where Korean stolen data is typically sold and requested help finding Korean Telegram groups used for data sales. In another session, the person asked Claude to draft a security-researcher résumé listing an age of 26, an education history, and a location in Maoming, Guangdong — details CrowdStrike believes likely belong to the attacker, though it could not establish that definitively.

The firm assessed with “moderate confidence” that the attacker is a Chinese speaker and financially motivated, rather than a state-directed operator. It has not attributed the campaign to a known hacking group or the Chinese government. CrowdStrike’s senior vice president of counter adversary operations put the significance plainly: the case shows how one human can target many victims in a very short period using AI agents.

Investigators also found that the intrusions exploited server-side authentication weaknesses in bank interfaces common to legacy systems — which is why the same method succeeded across multiple institutions. According to South Korean security firm Enki WhiteHat, the shared weakness let the attacker reuse the same approach again and again.

The warning nobody acted on

In a striking detail, South Korean outlet Aju Press reported that the sector’s security agency had walked bank security staff through a similar AI-powered attack weeks before the breaches — at a September 17 seminar attended by about 160 security officers. Staff from four of the firms later breached were in the room. In that earlier case, an attacker had told a DeepSeek-based agent to scan the financial sector for servers running Oracle WebLogic, exploited a flaw, and planted a web shell. The warning did not stop the September-October campaign.

President Lee Jae Myung has ordered an extensive investigation and stronger cybersecurity protections, and South Korean police have opened a probe. The Financial Services Commission says more than 68,000 people have been affected.

The tool’s developer pulls the plug

In a significant development, the developer of ARTEX has responded to the scandal by pulling the tool. Writing on GitHub, “Autumn-27” acknowledged the tool had been “abused by some bad actors,” saying such use was “entirely contrary to” their intentions. The project will no longer be updated and will be converted to closed-source: no further public versions, no maintenance support.

The move closes one door but highlights a bigger problem. ARTEX was one of a new generation of “agentic” AI tools — software that carries out chains of tasks on its own after a single instruction — and open-source code, once downloaded, cannot be recalled. Copies of ARTEX are almost certainly still circulating.

What it means for Pakistan

Pakistan’s banking sector should treat this as a live case study, not distant news. Pakistani banks run many of the same legacy interfaces and Oracle-based systems the attackers probed in Korea, and the country has already seen its own data breaches in recent years. The lesson from Seoul is that AI tooling lets a single attacker scale what used to require a whole team — probing many institutions in parallel for shared weaknesses.

For bank customers in Pakistan, the exposed data type matters most: names, phone numbers, and income figures are exactly what fuels SIM-swap fraud, phishing calls, and social-engineering scams. If you receive calls or messages quoting your personal details from someone claiming to be your bank, verify through the bank’s official channel — never through a number the caller provides.

The broader takeaway, echoed by security firms worldwide, is that defensive AI has to keep pace. Banks that still rely on manual security reviews and legacy authentication are facing attackers who automate both. Regulators, including Pakistan’s State Bank, will be watching how Seoul responds — and may be forced to mandate the same emergency hardening.

The bottom line

The South Korean bank hacks mark a turning point: AI-assisted cybercrime is no longer theoretical. A single motivated individual, armed with open-source AI agents and a mainstream coding assistant, breached institutions holding the financial data of tens of thousands of people in roughly two weeks. The ARTEX developer has pulled the tool, but the capability is out of the bottle. The question now is whether banks — and the people who trust them with their data — can adapt before the next campaign.