The region's risk this window was third-party operational. In Japan the compromised asset was a VPS platform that brokers and their customers trade on; in Korea it was internal web and groupware systems at institutions that disclosed separately and on different dates.
Japan — GMO Internet: data published after the ransom was refused
HighIn window
GMO's twelfth report on the Onamae.com Desktop Cloud intrusion, published 27 October 19:10 JST, confirmed that data claimed to be customer information had been posted, and stated the company will permanently refuse the demand — on the reasoning that payment would not guarantee deletion and would fund further crime. It has consulted police, is preparing legal action against parties spreading the data, and warned customers not to visit the leak site because visitor information was reportedly being harvested.
The timeline matters for anyone reading only today's headlines. Unauthorised access was detected on 20 October; the service was shut down on 21 October; the in-window event is the publication confirmation. Up to 26,739 customer servers may be affected. GMO told customers to back up, disconnect from the internet, power down and unplug cables before a threatened noon payload — the payload was never activated on customer machines.
The financial consequences are already visible at the edges. Japan's finance minister said five forex firms using the service — including OANDA Japan, Goldenway Japan and Ava Trade Japan — were affected and reported to the FSA. Customers have described unauthorised FX trades with losses in the low millions of yen. GMO is offering JPY 5,000 plus up to JPY 3 million per published item, and a bounty of up to USD 1 million for information leading to arrests. Authenticity of the leaked data remains unconfirmed by the company.
Sources: GMO Internet — 12th report, 27 Oct · NHK World — data released after ransom ignored, 27 Oct · Asahi Shimbun, 27 Oct · TechCrunch, 27 Oct · BleepingComputer — payload never activated · Kyodo — USD 1M bounty
Korea — Daishin Securities: 38,490 customers, announced today
High
Daishin Securities announced on 28 October that 38,490 customers were affected. Abnormal access was detected on 17 October on a server linked to remote account opening for discretionary investment, and confirmed on 26 October after reviewing more than 60,000 signals. Eight data types were exposed: names, emails, addresses, telecom identifiers, bank names, and encrypted account numbers, Daishin account numbers and mobile numbers. The company states there is no evidence of access to core systems, and its chief executive apologised and pledged compensation under law.
Sources: Asia Business Daily, 28 Oct · Korea JoongAng Daily, 28 Oct
Bithumb confirmed on 27 October that an unauthorised external party had abnormally accessed its internal mobile groupware system; the scope, including whether member data was involved, was still under analysis at publication. Separately, KB Kookmin Bank revised its affected population from 119 to 2,210 — 460 employees and 1,631 customers — on 27 October. That breach was detected on 30 September after an attack lasting roughly 43 hours and disclosed on 2 October, so the in-window element is the revision, not a new intrusion.
What we are not saying: that these share an attacker. The FSS is reported to be examining common vectors and possible AI use; that is an investigation, not a result. Mirae Asset Life's breach (26 October) and Mirae Asset Global Investments' blocked intrusion attempt sit just outside this window and are recorded as context only.
Sources: SBS News — Bithumb, Mirae Asset, Daishin, 27 Oct · Aju Press — KB Kookmin 2,210, 27 Oct · Yonhap, 28 Oct · SBS — FSS governor on AI-assisted attack, 27 Oct
Japan and Korea — the AI-agent reporting, and the three things it does not say
MediumActor unresolved
TrendAI's analysis, published in English on 27 October (Japanese original 23 October), separates three infrastructures. Case A used CyberStrikeAI, an open-source AI-agent penetration-testing platform, against at least 19 organisations, including at least 16 in Japan, plus Taiwan and Hong Kong — targeting legacy web systems at smaller businesses and local public-sector bodies. Case B used an unnamed LLM to accelerate planning and tool development, scanning at least 6,000 Japanese IP addresses with deep compromise at one or two companies over about 24 days. Case C used custom LLM-written scripts against at least 1,300 WebLogic candidates and at least 35 VPN entry points, including the Yamaha RTX1210 common in Japan.
Three deliberate omissions are as important as the findings. It names no financial institution among the victims. It withholds all IOCs to avoid interfering with investigations. And it states the three are distinct from ARTEX, that whether they share one operator is unresolved, and that Simplified Chinese being the working language in all three "is only a weak clue to location or affiliation and cannot establish either". Earlier ARTEX reporting (20–21 October, outside this window) remains the only current vendor chain that names finance as a target set in the region — approximately 80 Japanese companies targeted, eight breaches disclosed in finance, logistics and e-commerce, with a CrowdStrike-reported link to South Korean financial institutions in September.
No source links ARTEX or CyberStrikeAI to the GMO breach, and GMO has not established one.
Sources: TrendAI — Not only ARTEX, 27 Oct · Nippon.com — ARTEX, 20 Oct · CNA, 20 Oct · WSJ, 20 Oct