Are Banks Hiding Security Risks That Could Put Your Money at Risk? Inside the ARTEX Breach

On October 1, 2026, Shinhan Bank, one of South Korea's largest lenders, reported that an intruder had extracted personal data on about 25,000 customers from a loan-broker inquiry service, a count later put at 25,729 people [1][6]. Within days, six more Korean financial firms disclosed breaches of their own, and the official toll climbed past 68,000 [2][7].

The scale of the theft was not the unusual part. The tool was. Investigators at the Korea Financial Security Institute traced the intrusions to ARTEX, an open-source, AI-driven penetration-testing program developed in China, confirming that the attacker had used software that directs large language models to find weaknesses and work out how to exploit them [1][3][4]. The Wall Street Journal described it as one of the first AI-powered intrusions into the global financial system [2].

No accounts were taken over. Investigators said the attacker got in and pulled information out rather than seizing control, and that they do not expect funds to be moved out of customer accounts directly [1]. What was taken was the raw material of fraud: names, phone numbers, annual incomes, borrowing limits and, for some Shinhan customers, resident registration numbers, Korea's national ID numbers [1][3][8]. At least seven financial firms were hit, among them Shinhan, KB Kookmin, Hana and BNK Busan, two savings banks, and the consumer lender Hyundai Capital [2][8]. Two more banks, Woori and NH NongHyup, were attacked but not breached [1].

So, did the banks hide the risk? On this occasion, mostly not: they reported within days and disclosed how many were affected and what was taken. The case is still worth examining through that lens, because so much of what the public learned came not from the banks but from an investigator's phone call to a newspaper, a threat intelligence report, and documents a lawmaker pushed into the open [1][4][6]. Between those disclosures sit the things no rule requires a bank to tell you: which systems and outside technology companies touch your data, how often attackers come knocking, and what happens when the answer arrives too late.

What the ARTEX attacks actually did

ARTEX is not, on the available evidence, a state weapon. It is a penetration-testing platform, software built for the legitimate work of probing systems for weaknesses before criminals find them, distributed mainly through GitHub and aimed at Chinese-speaking users [1][3][7]. Released this year [7], it can scan systems for vulnerabilities, chain them into attack paths and try them without a human typing each step [3][8]. The tool was posted to GitHub on July 26, and its latest version appeared on September 24, three days before the first attacks on KB Kookmin began, according to accounts of the investigation [8]. It can plug into models from several providers, including DeepSeek, OpenAI and Anthropic [3].

The attacker did not have to be a sophisticated operator. CrowdStrike Intelligence found open directories on two servers, one in Hong Kong and one hosting an ARTEX instance likely used for the Korean attacks at 38.244.50[.]120 [4]. The exposed files included session histories from Claude Code, a command-line coding tool, along with ARTEX configuration files and memory files [4]. The ARTEX instance ran on DeepSeek v4.1-flash as its main model, accessed through what CrowdStrike assessed as a likely API reseller, with Zhipu AI's GLM-5.3 and Grok 4.6 used in separate sessions [4]. The logs show the attacker asking an AI assistant where stolen Korean data is typically sold and for help finding Korean Telegram groups for data sales [4].

CrowdStrike assessed with moderate confidence that the attacker is a Chinese speaker motivated by money, and it did not attribute the activity to any known group [4]. The attacker's own carelessness narrowed the field anyway. In one session, the attacker asked Claude to write a security researcher's résumé touting the campaign's results, complete with a name, a Telegram handle, an age of 26, a South China University of Technology education and an address in Maoming, Guangdong [4][5]. CrowdStrike cautioned that those details cannot be definitively tied to the person behind the intrusions [4][5]. Korea's investigators were blunt about who was in charge. The AI was used in the attacks, but "the AI did not act independently without human involvement," an institute official said. "A hacker used the AI as a tool" [1].

Every intrusion followed the same shape, and none came through the front door. The targets were internal, employee- and partner-facing systems: Shinhan's loan-agent inquiry service, KB Kookmin's employee mobile work-support system, Hana's sales-support system [1]. Customer-facing internet and mobile banking, the side whose security had been strengthened the most, was left alone [1]. Investigators described automated attacks feeding large numbers of guessed customer numbers into these peripheral systems until valid records came back, with two to three IP addresses overlapping at each firm and the attacker switching whenever one was blocked [1][8]. "Blocking IPs is little more than emergency first aid," the official said [1]. What tied the campaign together was a recognizable signature in the traffic, which is how investigators identified ARTEX and linked the firms [1].

By recent Korean standards the breach was small: nearly 40 million users of the streaming service Tving were exposed in June, and 33 million Coupang users last year [3]. The speed was the new part. "It would take weeks, if not months, for human hackers to plan and carry out such large-scale attacks," said Lim Jong-in, a professor at Korea University's Graduate School of Information Security, who compared the difference to that between a person running and a Ferrari [3].

Was customers' money taken?

Not in this case, according to investigators. The institute's assessment was that direct financial loss is unlikely, because the attacker "did not take over the systems" but got in and extracted information by querying them [1]. The breached systems were supporting networks, not the rails that move payments [3].

What the attacker took, firm by firm:

• Shinhan Bank: 25,729 people, exposed through a loan-agent inquiry service, including names, phone numbers, annual incomes, loan limits and, for some, resident registration numbers [1][3][8]
• Yegaram Savings Bank: about 40,000 customers [6][8]
• Welcome Savings Bank: up to 2,200 corporate clients [8]
• Hyundai Capital: 146 housing-loan agents [8]
• KB Kookmin Bank: 119 records, from an employee mobile work-support system [1]
• Hana Bank: 89 records, from an employee sales-support system [1]
• BNK Busan Bank: 11 contract workers [1]

The risk to customers lies in what those fields allow. Annual income, a borrowing limit and a phone number add up to a script for a convincing call: the person on the line knows what you earn, what you could borrow and which bank to name. Regulators warned that the leaked details hand voice-phishing gangs exactly that [8], and the institute said the data could be used for voice phishing and similar scams [1]. Identity theft is a second channel, since national ID numbers were exposed for some Shinhan customers [8].

The money that did move was moved by customers themselves. Demand deposits at the five largest banks fell from 675.5 trillion won on September 30 to 655.4 trillion won, a drop of just over 20 trillion won, about $15 billion, in 11 days, according to figures reported by the Korea Times [6]. No accounts had been emptied. Trust had been.

What the seven lenders disclosed, and when

Shinhan reported first, on October 1 [1][6]. Yegaram had already disclosed through a notice on its website that an unidentified hacker gained access [1]. The National Police Agency's cyber terror unit opened an investigation, announced on October 6 [2]. The Korea Internet & Security Agency raised its threat level from "interest" to "caution" [8], and the Financial Services Commission put the sector toll at more than 68,000 people [7].

The disclosures were prompt by global standards, but the full extent took days to establish, and the reason is the most revealing part of the story. Several banks did not know they had been breached until after Shinhan reported. The institute shared the attacker IP addresses identified in its Shinhan investigation with the wider financial sector, and only then did other banks review their access logs and, in the official's words, belatedly discover and report breaches of their own [1]. The confirmed count stood at just under 26,000 on October 3 [1]. It passed 68,000 as breaches at further firms, including the two savings banks, were confirmed [7][8]. For days, stolen data sat with the attacker while some institutions still did not know they had been hit. Nothing in the public record shows customers were harmed during that window. The window is still the point: notice rules work at the speed of detection.

The regulatory response was sector-wide. President Lee Jae Myung ordered a thorough investigation on October 4, and Financial Services Commission Chairman Lee Eok-won chaired an emergency meeting of financial-sector chiefs the same day [8]. "Financial security is the foundation of trust and stability," Lee said. "We must review the entire information security system from square one" [8]. The commission ordered every financial institution to review security from scratch: block external access by default unless a service requires it, inventory and test every internet-facing system, cut employee access rights to the minimum, share attack indicators across the sector, and adopt AI-based monitoring that can match the speed of automated attacks [8]. Officials also floated easing Korea's strict network-separation rules so banks can run modern defensive tools, a reversal of the long-held view that air gaps alone keep core systems safe [8]. The FSC additionally moved to recommend comprehensive audits of internal employee-facing systems [1]. Shinhan pledged compensation to affected customers [8].

What the banks disclosed themselves, in notices and briefings, was the practical core: how many people were affected, which systems were hit and what data was taken [1]. What they did not disclose, and may not have known, was the AI dimension. That entered the public record on October 3, through the institute's confirmation to the Herald Business, two days after Shinhan's first report [1].

Where this story sits

5 related pieces

Hover a story · click to read

1. No. 73 · Oct 20, 2026After ARTEX: How the PRA, FSC Korea and the ECB SSM Diverged on AI Bank Risk
2. No. 71 · Oct 17, 2026Seven Institutions After ARTEX AI: What South Korea's Banks Changed, and What's Still a Promise
3. No. 61 · Oct 7, 2026ARTEX AI Hit Seven South Korean Banks. Attribution Just Got Harder.
4. No. 64 · Oct 11, 2026The ARTEX AI bank hack claim: what South Korea's record actually shows
5. No. 3 · Aug 25, 2026When the Money Runs Out: The Economics of the Breaking Point

The warning the banks received two weeks earlier

The most consequential fact surfaced last. On September 17, the institute briefed about 160 bank security officers on an AI-agent attack against a domestic lender it called Company A, according to documents it later submitted to National Assembly lawmaker Han Chang-min, disclosed on October 11 and reported by the news agency AJP [6]. The case was concrete. An attacker directed an agent built on the Chinese model DeepSeek to scan the financial sector for servers running Oracle WebLogic. The next day the agent exploited a remote command execution flaw on one of those servers and planted a web shell. Company A caught the intruder just before more malicious code could be pulled onto its internal servers [6].

The institute's advice to the room was direct: assessments done once or twice a year will not catch this, and internet-facing systems need constant measurement and verification [6]. Security staff from four of the firms later breached were in attendance [6]. Two weeks later, the same class of attack, run through a tool built to automate exactly that workflow, hit seven firms in their sector [2][6]. The seminar did not become public until a lawmaker pushed the documents out nearly a month later [6].

The investigation remains open in the latest public accounts. As of October 3, the institute was still analyzing two savings banks' logs to confirm breaches, officials said far more institutions were attacked than were breached, including organizations outside banking, and there was no clear timeline for completing the probe [1]. The government assessed that ARTEX was highly likely to have been used at more than seven institutions, and researchers cautioned that the number of breaches confirmed to involve the tool remained unresolved [7]. The attacker's identity, too, remained unconfirmed in public reporting as of mid-October [4][5].

Where the disclosure rules fall short

Start with what Korean rules required. Institutions had to report incidents that caused damage; when none occurred, they need not report at all. "Financial institutions are not required to report incidents when no damage occurs," the institute official said [1]. Woori and NH NongHyup repelled the attacks; no rule required them to say anything, and the public learned of those attempts through the institute official's account [1]. On the consumer side, the Personal Information Protection Act, amended in 2023, requires companies to report leaks that cause actual harm to the privacy regulator within 72 hours and to notify affected individuals without delay when damage occurs or a leak involves at least 1,000 people [15].

On AI and outside vendors, the gap is wider still. Korea's rules, like most of their counterparts, attach disclosure obligations to incidents, not dependencies. Nothing in them required the banks to tell customers which AI systems, cloud services or third-party providers touch their data, or to warn them when those dependencies create risk. The regulator filled part of that hole in its October 4 order, but its requirements run between banks and supervisors, not from banks to depositors [8].

The same asymmetry runs through the other major regimes. In the United States, the SEC's 2023 cybersecurity rule requires public companies to disclose a material cybersecurity incident on Form 8-K within four business days, but the clock starts only when the company itself decides the incident is material, and the annual Form 10-K item describes processes and past incidents rather than a live picture of current risk [9]. The OCC, the Federal Reserve and the FDIC require banks to manage third-party providers as if the outsourced work were the bank's own [10], but that oversight happens between banks and examiners. The EU's Digital Operational Resilience Act, in application since January 2025, forces banks to keep registers of their technology contracts, report major ICT incidents to supervisors and submit designated critical providers to direct oversight [11]. Those registers go to regulators, not to the public. The United Kingdom folds AI into existing rules on outsourcing, operational resilience and individual accountability, following the Bank of England and Financial Conduct Authority's joint review of the technology [13]. The Financial Stability Board, in its November 2024 report on AI in finance, flagged exactly the dependency this case exposed, concentration in a small set of cloud and model providers, and found most jurisdictions relying on existing frameworks rather than writing AI-specific disclosure rules [12].

The result is a one-way mirror. Defenders can consult MITRE's ATLAS, a public knowledge base of adversary tactics and techniques against AI systems, drawn from real-world cases [14]. Customers have no equivalent register of the AI systems and vendors their bank depends on. In most countries, the only guaranteed disclosure about a bank's technology dependencies is the one that arrives after something has gone wrong.

Did the banks' own AI cause the breach?

The record does not show that. The breached systems were conventional internal applications: a broker portal, a staff mobile app, a sales-support tool [1]. The AI was on the attacker's side of the connection. The failure on the banks' side was older and simpler: an uneven perimeter. "Security on customer-facing electronic financial services has been enormously strengthened, but internal employee-facing systems had been managed less rigorously," the official said [1]. Hwang Suk-jin, a professor at Dongguk University, put it more plainly: the attackers "came in through a window," targeting supporting networks with weaker defenses rather than the core [3].

The AI matters for a different reason: tempo. A single operator, using a tool that automates reconnaissance and exploitation, probed an entire banking sector in a matter of days [3][8]. Lim Jong-in cautioned against assuming the attack was state-directed simply because the tool was Chinese-language; Chinese models with relatively weak safeguards can be downloaded and modified by anyone [3]. Hwang raised a second concern, that the breach may have been an initial test, with larger follow-up attacks possible using the information obtained [3]. The institute itself said attacks of the same type could be repeated at any time, and noted that there are many open-source AI tools like ARTEX [1].

The developer, who posted an October 8 statement under the GitHub handle Autumn-27 and was identified in reporting as Li Puhua, a Chinese cybersecurity engineer whose tool had won a Chinese agentic AI competition in September, said ARTEX had been misused, disavowed involvement and condemned unauthorized use, and announced that updates would stop and the project would go closed source [7]. The withdrawal has limits. Copies already downloaded keep working, and researchers have found English- and Korean-language derivatives of the existing code [5][7]. AhnLab, a Korean security firm whose intelligence unit first published findings on ARTEX on October 1, identified ARTEX consoles exposed on 394 systems in an infrastructure survey, an indication of how widely the tool had spread, though exposed consoles do not by themselves show malicious use [7].

The banks, meanwhile, are turning to the same class of technology for defense. NH NongHyup plans an AI red-teaming system, KB Kookmin intends to lift its security budget above 100 billion won next year from 86.1 billion won, and Woori is weighing an increase of more than 20 percent [6]. Shinhan, KB and Hana together spent about 124 billion won, roughly $92 million, on information security last year [8]. The heads of the five largest banks were called to testify at a National Assembly audit on October 19 [6]. The open question for the sector is whether the next tranche of spending reaches the windows.

What protections customers actually have

Against the bank, Korean law gives notification and damages. Firms must notify affected individuals when a leak causes damage or involves at least 1,000 people [15], and the Personal Information Protection Act allows breach victims to seek compensation from a company whose intent or negligence contributed to a leak [15]. Shinhan has pledged compensation to affected customers [8].

Against fraud, Korea has a reimbursement rule. The Electronic Financial Transactions Act, amended in 2024, requires banks to partially reimburse victims of electronic financial fraud such as voice phishing, with payouts capped at 30 million won and the share depending on how quickly the victim reports [16].

Against collapse, deposit insurance works the same way in Korea and the United States: it protects your deposits if the bank itself fails. It does not cover money you were tricked into sending somewhere else.

The strongest protection in most consumer regimes is speed. Under the U.S. Regulation E rule governing electronic transfers, liability for an unauthorized transfer is capped at $50 if the customer notifies the bank within two business days of learning of the loss or theft of a card or other means of access, at $500 if the report comes later, and losses that occur more than 60 days after the statement showing an unauthorized transfer can fall on the customer [17]. The rules reward the customer who reports fast, which is precisely what a scammer armed with your income and borrowing limit will try to prevent by manufacturing urgency.

How to protect your money and personal information

The immediate threat from this breach, as investigators framed it, is not an empty account; it is a convincing phone call [1]. The defenses are ordinary but specific:

• Treat a caller or message that already knows your annual income or loan limits as a red flag, not a credential. Those are among the fields the attacker extracted [1][3].
• No bank will ask you to read out a one-time passcode, move money to a "safe account" or verify your credentials on an inbound call. Hang up and call back on the number printed on your card or in the official app.
• Turn on transaction and login alerts, and read your statements. Timing decides how much protection you get under rules such as Regulation E [17].
• If your institution notified you, follow its compensation process. Shinhan has pledged compensation to affected customers [8].
• Change passwords you have reused, and watch for accounts or loans opened in your name. National ID numbers were exposed for some Shinhan customers [8], which raises identity-theft risk beyond ordinary phishing. Check your credit record; in the United States, a credit freeze at each of the three bureaus is free.
• Report phishing to your bank and to the police. Korea's National Police Agency cyber unit opened an investigation into the breaches [2].

One expectation worth calibrating: the attacker kept switching IP addresses while the method stayed the same [1]. Banks can block addresses. Only you can refuse to be talked out of your money.

Why this will happen again

The honest answer to the headline's question is that, on this occasion, the banks did not hide the breach. They reported within days, disclosed the counts and categories, and one pledged compensation. What stayed hidden were the things no rule forces into the open: that the industry had been shown a live AI attack two weeks before the breaches [6]; that several banks learned of their own intrusions only when investigators handed them the attacker's addresses [1]; that far more institutions were attacked than were breached, including firms outside banking [1]; and that the tool behind it all was a free download that still exists in downloaded copies and derivatives [5][7].

Detection, not disclosure, was the weak point. The campaign was unmasked partly because the attacker left session histories and configuration files in open directories on the attacker's own servers, giving researchers a direct view of the operation [4][5]. The next attacker may not be so careless. The next bank may not get a warning seminar, a shared address or a journalist's phone call.

Regulators are writing rules for what this case exposed: technology-dependency registers in Europe, incident clocks in the United States, third-party oversight on both sides of the Atlantic, and in Korea a sector-wide order to inventory what faces the internet [8][9][10][11][12]. Nearly all of it flows to supervisors rather than to you. Until a rule requires banks to tell customers which systems, models and vendors hold their data, and what happens when those dependencies fail, the practical rule for customers is the inverse of the bank's: assume the perimeter is uneven, treat unexpected knowledge of your finances as a threat, and never let urgency decide where your money goes.

Sources

1. Jeong Ho-won, "Exclusive: Chinese AI tool ARTEX used in wave of bank hacks, probe finds," The Herald Business, October 3, 2026. https://biz.heraldcorp.com/article/10892497
2. Sooyoung Rhee and Raffaele Huang, "Hackers Use Chinese AI Tool to Hit South Korean Banks, Exposing New Risk," The Wall Street Journal, October 6, 2026. https://www.wsj.com/world/asia/hackers-use-chinese-ai-tool-to-hit-south-korean-banks-exposing-new-risk-5d4d3885
3. Lee Hyo-jin, "[EXPLAINER] How AI emerged as new threat in Korea's bank hacking crisis," The Korea Times, October 7, 2026. https://www.koreatimes.co.kr/business/banking-finance/20261007/explainer-how-ai-emerged-as-new-threat-in-koreas-bank-hacking-crisis
4. CrowdStrike Intelligence, "Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance," October 7, 2026. https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/
5. Bill Toulas, "ARTEX AI, Claude agents used in cyberattacks on South Korean banks," BleepingComputer, October 10, 2026. https://www.bleepingcomputer.com/news/security/hacker-used-artex-ai-and-claude-agents-to-target-south-korean-banks/
6. "Korea's Banks Were Shown the AI Attack Two Weeks Before It Hit Them," aravind-r.com, October 12, 2026. https://aravind-r.com/blog/korea-banks-ai-attack-warning-artex-breaches
7. "ARTEX ends public releases after South Korean bank attacks," Magica, October 9, 2026. https://magica.com/news/artex-developer-closes-source-after-south-korean-bank-attacks
8. "Korea Probes AI-Tool Hacks at Seven Lenders, 67,000 Customers Hit," Santage, October 5, 2026. https://santageai.com/news/2026/10/05/korea-banks-artex-ai-hacking-probe
9. U.S. Securities and Exchange Commission, "SEC Adopts Rules on Cybersecurity Risk Management, Strategy, Governance, and Incident Disclosure," press release 2023-139, July 26, 2023. https://www.sec.gov/news/press-release/2023-139
10. Office of the Comptroller of the Currency, "Third-Party Risk: Interagency Guidance on Risk Management of Third-Party Relationships," OCC Bulletin 2023-17, June 2023. https://www.occ.gov/news-issuances/bulletins/2023/bulletin-2023-17.html
11. Regulation (EU) 2022/2554 on digital operational resilience for the financial sector (DORA), Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2022/2554/oj
12. Financial Stability Board, "The Financial Stability Implications of Artificial Intelligence," November 2024, fsb.org.
13. Bank of England and Financial Conduct Authority, "Artificial intelligence in financial services," Discussion Paper DP5/22, October 2022, with associated 2023 feedback statements, bankofengland.co.uk.
14. MITRE, "ATLAS: Adversarial Threat Landscape for Artificial Intelligence Systems." https://atlas.mitre.org/
15. South Korea, Personal Information Protection Act, as amended 2023.
16. South Korea, Electronic Financial Transactions Act, as amended 2024.
17. United States, 12 C.F.R. § 1005.6, Regulation E, Electronic Fund Transfers.