Tokyo’s Roppongi district and Tokyo Tower at night, seen from Shibuya, where GMO Internet Group is based (illustrative). Image: Syced / Wikimedia Commons, CC0, cropped
The Chinese AI hacking tool suspected in this month’s raids on South Korean banks has turned up behind a wave of attacks on Japanese companies. GMO Cybersecurity by Ierae, the security arm of Tokyo’s GMO Internet Group, said on October 19 that it had confirmed ARTEX, an AI system that runs penetration tests on its own, was in use on the infrastructure behind recent break-ins at Japanese firms.
Japan has had a run of breach disclosures since September, from restaurant chains to a car hire operator. GMO says at least some of them trace back to the same attackers, and that those attackers were handing much of the work to AI.
One server, about 20 Japanese targets #
GMO, which says it investigated with outside organisations at the request of affected companies, started with one attack server. Its targets were almost all in Japan: about 20 companies, nine of them sites in the same industry, a pattern GMO reads as a campaign aimed at particular countries and sectors. ARTEX was running on it.
It then found six more AI-driven attack servers aimed almost entirely at Japanese companies, each focused on a cluster of industries: retail and digital gifts, education and printing, marketing and loyalty points, insurance and finance, e-commerce, media and logistics, and e-commerce alone. Eight of the targeted firms had already announced data leaks in September and October, which GMO says points to “multiple incidents of damage” caused by the same group. It does not name them.
GMO is careful about what that means. Being targeted is not the same as being breached, it says, and it has not said who runs the servers or where they are. The Yomiuri Shimbun, which interviewed the company, reports that the six servers sat overseas, were active between September 18 and October 18, all ran ARTEX and may have been used against at least 80 organisations. Japan’s national computer emergency team, JPCERT/CC, told the paper it had also seen signs of AI agents in attacks on Japanese companies.
In an earlier post on October 14, GMO said forensic logs from the break-ins made it almost certain AI had been used, and explained why that matters: an AI can take a published vulnerability, trace it through every version of an application’s code and find copies of the same flaw that were never flagged.
“Almost exclusively targeting Japan” #
Daiki Fukumori, who heads GMO’s cybercrime countermeasures centre and wrote the post, said the company had passed its findings to law enforcement “to prevent further damage”. He told the Yomiuri that victims had “their minor system weaknesses targeted”, and had a warning for everyone else:
Companies should conduct a thorough review to ensure there are no overlooked vulnerabilities in their systems.
Daiki Fukumori, GMO Cybersecurity by Ierae, to the Yomiuri Shimbun
Chief Cabinet Secretary Minoru Kihara told reporters on October 20 that the government was working out who was behind the attacks and where they came from, considering all possibilities, including foreign involvement. The government has urged companies to strengthen their defences.
The developer pulled the plug #
ARTEX was posted publicly on GitHub in July by a Chinese-speaking developer, and it was already under suspicion in Korea, where four banks leaked data in two days and CrowdStrike later found an attacker’s ARTEX server, its configuration and Claude Code instruction files left in open folders. Hackers who may have used AI agents also took data on nearly a million members of a Seoul church.
On October 8 the developer said ARTEX would go closed-source, with no more public releases or maintenance. The tool was meant for authorised security testing, the developer said, and its use in cyberattacks was “entirely contrary to” those intentions. The GitHub repository now returns a not-found page, but copies made while it was public may still be out there.
It is not the only case of attackers leaning on AI this month: one hacker set AI agents loose on 1,200 websites and broke into Deere and SharkNinja.
Why it matters #
A free, open-source AI tool has now been tied to attack campaigns in two countries within a month, aimed at the small, forgotten weaknesses that human attackers rarely have time to hunt for. Pulling the code does not pull back the copies, and Japan’s government is still working out who used it.