South Korea’s president says “signs have emerged” that artificial intelligence was used to break into financial firms. Investigators have not confirmed it. What they have confirmed, according to the reports reviewed here, is that personal data on tens of thousands of customers is out, and that, by one account, the likeliest way in was neither new nor clever.
That combination is why the case matters beyond Seoul. If a patchwork of old tricks can be run faster and cheaper by software agents, every company with a forgotten login page is on the clock.
What is confirmed, and what is not
The incidents first surfaced on September 30, according to The Record, a cybersecurity news outlet. By early October at least seven financial institutions had reported leaks. Quartz, citing wider reporting, puts the number of people whose data was exposed at about 68,000, and names Shinhan Bank and KB Kookmin Bank among the affected firms. The Record also names Hana Bank.
Counts differ by outlet. A secondary analysis site, ExplainX, cites the Korea Herald for roughly 66,000 individuals plus about 2,200 corporate records. Another, Tech-Insider, says other domestic outlets reported more than 65,000 records. These totals are still being reconciled, and the second-hand accounts are not a substitute for official figures.
Shinhan Bank said about 25,000 of its customers were affected, according to The Record. The exposed information includes names, phone numbers, income and borrowing history, the Korea Herald reported, as relayed by The Record.
What the officials actually said
President Lee Jae Myung told officials that AI models appear to have been used and ordered an investigation, Quartz reports. The Record quotes him saying AI makes it possible to hack with ease even without specialized skills, and says he pledged staff and money to limit the damage.
The financial regulator’s tone is more careful. Tech-Insider, citing SBS News, reports that the Financial Services Commission chairman said AI use could not be ruled out. That is a different statement from saying it happened.
On the investigation side, The Record reports a 28-person team at the National Office of Investigation. Financial authorities reportedly traced 33 IP addresses linked to the attacks, and the Financial Supervisory Service said they connect to at least 12 countries, including Japan, the U.S., Thailand, Vietnam and Hong Kong. ExplainX, citing the Korea Herald, gives 28 IP addresses instead. The numbers conflict, and the differences may reflect when each count was made.
Regulators also warned that an IP address shows where traffic came from, not who was behind it.
The tool investigators are looking at
The Record reports that officials believe a Chinese-developed tool called ARTEX AI was involved. Tech-Insider, citing an analysis by the security firm Rescana, describes ARTEX as an open-source, LLM-based penetration-testing framework. Penetration-testing tools are built for security teams to probe their own systems, and this one can reportedly run reconnaissance, find login endpoints, try stolen or guessed passwords and check whether they worked, with limited human direction.
ExplainX, which we treat as a single secondary source, adds several details that other outlets have not confirmed in the reporting we read. It says a web server believed linked to an attack on Shinhan carried a Chinese-language page title reading “AI autonomous penetration testing console.” It also says CrowdStrike published a report on October 7 that assessed with moderate confidence that the actor is likely Chinese-speaking and financially motivated, and did not attribute the activity to the Chinese state.
Be careful with those details. A page title only shows that a dashboard was run or copied. It says nothing about whether an AI model picked targets or wrote exploits, and a tool name found on attacker infrastructure does not identify who used it.
The old trick underneath
According to ExplainX, the main method was credential stuffing. That means taking usernames and passwords leaked in earlier breaches and trying them against other sites, betting that people reuse them. The same account says the targets were side-door systems such as broker portals and employee work-support systems, not the banks’ main customer apps.
Tech-Insider’s analysis lands in the same place. Credential stuffing, endpoint scanning and IP rotation are all old techniques, and the new claim is that AI chained them together with less manual oversight. The same piece warns that AI fingerprints are hard to tell apart from skilled human automation.
Put plainly, AI may not have invented a new attack here. It may simply have made an old one cheap. Testing thousands of logins across dozens of forgotten portals used to need a team, and if that account holds, one operator with a script that adapts could now do the same job.
What it means for everyone else
Nothing here suggests ordinary customers did anything wrong. The likely lesson reaches well beyond banks, though, and most of it is plain security hygiene:
- Reused passwords are what make credential stuffing work. A password manager lets you use a different one everywhere.
- Put a second factor on every login page, including the dull ones: partner portals, vendor logins and internal tools.
- Watch behavior, not just addresses. The Korean attackers reportedly rotated IP addresses to dodge simple blocks, so rate limits and login monitoring have to look for patterns.
- Keep a list of every login page you expose to the internet. A system nobody remembers cannot be protected.
None of this is specific to AI, which is the point.
What to watch next
Several questions remain open, according to the reports reviewed. The final victim count is unsettled, and so is whether anyone has been charged. Which AI models, if any, directed the attacks has not been established by officials. ExplainX mentions parliamentary audits later this month, though the dates it gives conflict. Tech-Insider’s author expects the exposure totals to be revised, probably upward, over the next few weeks. That is a forecast, not a reported fact.
Editor’s take
Whether AI “did” this may never be settled, and I would not wait for the answer. Credential stuffing worked before AI, and it still works wherever a login page has no second factor or a team has forgotten a portal exists. I use AI tools every day for the dull, repetitive parts of my work, so it is easy to believe attackers do too. Check your old doors before you buy a new AI shield. That is my judgment as editor, not a finding of the investigation.
