GMO Internet Cloud Breach: What Is Confirmed and What Is Claimed About ARTEX

GMO Internet Group has confirmed that attackers breached its Onamae.com Desktop Cloud, a rented Windows server service that retail foreign-exchange traders use to run automated trading tools, and then used stolen credentials to place unauthorized orders at brokerages. The company says up to 26,739 servers could have been affected. The AI angle comes from the group's CEO, who reportedly pointed to an autonomous AI penetration-testing tool called ARTEX. That part needs careful reading, because the public record supports a narrower claim than some headlines do.

This post separates what GMO and the press have confirmed from what is only claimed, lays out the timeline, and ends with what customers and engineers can do. It sits next to our earlier coverage of Japan's run of cyberattacks and our running October 2026 attacks tracker.

Update (October 26, 2026): CEO admits known vulnerabilities, shares drop, cabinet-level talks reported

New details arrived on Monday, October 26. Treat them by source strength:

• Market reaction. Investing.com reports that GMO Internet shares fell 13.6% to 3,256 yen on Monday, while the broader TOPIX slipped 0.8%. Other outlets cite slightly different drops, so the exact figure depends on the time of day.
• Vulnerabilities flagged earlier. The same report says CEO Masatoshi Kumagai apologized on the evening of October 25 and acknowledged that internal engineers had identified security weaknesses that were not adequately addressed. That is a notable shift from the AI-tool framing: it points back to ordinary patching and remediation speed. We have not seen the original statement, so the wording is relayed.
• Compensation. Every user of the affected service gets a 5,000-yen goodwill payment, and the company has pledged to cover verified losses from unauthorized trades, matching the fourth notice described below.
• Government. The same report says the group petitioned the Japanese government to strengthen national cybersecurity defenses. Social-media reports of a cabinet-level meeting on the breach are unconfirmed, and we found no wire or government source for them.

Still missing: a root-cause disclosure, final loss totals, and proof of whether ARTEX performed this specific intrusion.

Update (October 27, 2026): reports of leaked data and a $50M demand are not confirmed

On October 27, an aggregator headline said hackers leaked stolen data after GMO Internet rejected a $50M ransom. We could not trace either detail to GMO or a primary outlet. GMO's notices as relayed by The Next Web say the attackers demanded payment and threatened unauthorized orders at more than 70 securities and FX firms, data leaks and ransomware. GMO said it will not pay, asked police to investigate, and warned that anything published may be partly false. BleepingComputer reports malware on up to 26,739 servers but gives no dollar figure.

Treat the $50M amount and the leak as unverified until GMO or police confirm. If you used the service, change passwords and end active sessions, and treat any leaked-data claims with caution.

Update (October 28, 2026): data posting and the roughly 8 billion yen demand are now reported, plus a $1M bounty

Two items from the previous update have firmed up. TechCrunch reported on October 27 that GMO said on its own blog it believes the attacker has published information allegedly tied to customers and the business, while the company is still checking whether it is genuine. TechCrunch's timeline also says GMO received an email on October 21 demanding about 8 billion yen (roughly $50 million), which matches the figure we had flagged as unconfirmed. GMO also told customers to reset passwords after reports of unauthorized operations on trading accounts, and TechCrunch notes the company has not publicly established a link between ARTEX and the Onamae breach.

On October 28, The Hacker News reported that GMO is offering up to $1 million for tips that help catch the attackers. Other search snippets say the payout applies only if a tip leads to an arrest or an international wanted notice, and that the attackers lowered their demand to about 160 million yen. We could not open the article or find GMO's own notice for these two details, so treat the bounty terms and the reduced demand as reported, not confirmed. Be careful not to mix up the numbers: the ransom figures and the $1 million reward are separate things.

Quick answers: confirmed vs claimed

What is Onamae.com Desktop Cloud, and why did the breach hit brokerage accounts?

According to a translated analysis from Lilting, which draws on GMO's own notices, Onamae.com Desktop Cloud rents Windows virtual private servers so customers can run FX auto-trading tools such as MetaTrader 4 and 5 around the clock. Those tools store the login ID and password for each customer's brokerage account. The servers therefore hold the keys to accounts that live at brokers and FX firms, not at GMO.

That design explains the blast radius. Shutting down GMO's servers did not stop an attacker who already held the credentials from placing orders from elsewhere. It is the same lesson we keep seeing with agent and automation platforms: a convenience layer that stores long-lived secrets becomes the target. Our coverage of AI agents moving into financial accounts makes the same point from the bank side.

Timeline from GMO's notices

The dates below come from the Lilting translation of GMO's fourth and fifth notices. All times are JST.

The first notice had said GMO had not confirmed misuse of the viewed information. The second notice changed that, which is worth remembering when judging early incident statements: they are often revised within days.

What did the attackers demand?

GMO disclosed that people believed to be the attackers demanded money. As relayed in the Lilting translation, the threats were to keep placing unauthorized orders at more than 70 brokers and FX firms in Japan and abroad, to publish information about the service and about customer PCs that had connected to it and carry out a ransomware attack, and to publish information about GMO's systems and operations. GMO said it would not pay and handed communication records to the police cybercrime unit.

One customer interviewed by TBS saw a balance drop to zero on October 21, with losses of more than 4 million yen within two seconds. A sender calling itself "HackingTeam" then emailed that customer, per TBS as summarized by Lilting. A claim-of-responsibility site reportedly said the group had destroyed accounts worth about 7.7 billion yen in 24 hours. That is the attacker's claim, and the real damage has not been confirmed.

What is the ARTEX claim, exactly?

There are two separate threads, and they should not be merged.

Thread one: GMO Cybersecurity's finding (October 19-20). Jiji Press via Nippon.com reported that GMO Cybersecurity by Ierae, a GMO Internet Group subsidiary, traced cyberattacks on Japanese companies to six servers, all of which used Artex, an AI-powered penetration-testing tool. The attacks from those servers targeted Japanese financial and logistics companies, eight of which disclosed information breaches between September and October. GMO Cybersecurity said multiple incidents were caused by the same group. Jiji also reported that Artex is believed to have been used against South Korean financial institutions. Other outlets describe ARTEX as a Chinese-developed, LLM multi-agent tool; those secondary details come from press coverage we could not open directly, so treat them as reported.

Thread two: the CEO's statement (October 24-25). Secondary reports, such as this one from OfficeChai, say Masatoshi Kumagai blamed misuse of ARTEX and warned about AI misuse. We could not open that report or the original interview, and in our searches no readable source tied ARTEX to the Onamae.com Desktop Cloud intrusion. We also found no public GMO notice detailing how the intrusion began. Treat the CEO link as unconfirmed.

So the careful statement is this: GMO's security subsidiary has publicly tied an AI pentesting tool to attacks on Japanese firms, and the group's CEO links the broader wave to AI misuse. What is not yet public is evidence that ARTEX performed the specific intrusion into Onamae.com Desktop Cloud, or which vulnerability was used. Sources do not yet agree on loss totals, and the 26,739 figure is a potential count.

Why this matters beyond Japan

An autonomous pentesting agent is a dual-use product. The same capability that lets a defender find weaknesses lets an attacker scale reconnaissance and exploitation. This is the pattern behind AI agents targeting Latin American financial firms, and it fits warnings from Bill Gates about AI-driven cyberattacks and the G7 finance tabletop exercises on AI cyberattacks.

It also shows the limit of an "AI did it" explanation. Whatever role ARTEX played, the design stored credentials where one compromise exposed everything. AI raises the tempo; basic patching and secret hygiene still decide the outcome.

Japan's active cyber defense: what GMO asked for

On October 24 GMO asked the government to apply "active cyber defense" and called the incident the first activation case. Per Lilting, the framework lets police and the Self-Defense Forces access servers used in attacks and neutralize them, and the relevant part took effect on October 1, 2026. Neutralization means deleting malicious programs or changing settings on a server, not physically destroying it.

The police clause does not require the attacker to be a state, and it covers serious harm to people's property when there is an urgent need. But there is no procedure for a private company to apply; GMO's request is input to a government decision. For servers abroad, the National Security Council reviews a response policy, the Foreign Minister is consulted, and an independent oversight commission approves in advance. As of 17:00 JST on October 24, no government response had been found. Neutralization also cannot restore balances already lost; losses are handled through GMO's compensation and the brokers.

What GMO is offering customers

Per the fourth notice as relayed by Lilting: a flat 5,000-yen payment to every current user of the service, compensation for realized losses from unauthorized trades, and six months of free use of a replacement service on a separate cloud platform. GMO asks customers to contact their brokers about cancelling trades or getting refunds, and says it will pay whatever remains after any refund.

How to protect yourself

If you used the service or a similar hosted trading setup:

1. Change the brokerage password, then terminate all active sessions or ask the broker to disconnect them. GMO warns earlier connections may remain active after a password change.
2. Enable multi-factor authentication and API or IP restrictions where your broker offers them.
3. Review recent trades and contact your broker promptly about disputed orders.
4. Treat any machine that connected to the affected service as potentially exposed and rotate other credentials stored on it.
5. Be wary of emails claiming to be from the attackers or from GMO; verify through official notices.

If you run cloud or agent infrastructure:

• Close known findings quickly. An unpatched vulnerability is the cheapest way in.
• Do not keep long-lived brokerage, payment or admin credentials on a shared server; use scoped, revocable tokens.
• Assume attackers have automation. Monitor for rapid, scripted actions such as many orders within seconds.
• For AI coding and operations agents, put guardrails on what they may do. AgentBeam, the agent security platform from the explainx.ai team, stops AI agents before they take dangerous actions.

What to watch next

• A GMO root-cause disclosure naming the vulnerability and whether ARTEX was used on this intrusion.
• Final loss figures and how many of the 26,739 servers were actually compromised.
• Whether the government activates neutralization, and against servers in which country.
• Whether the developer of ARTEX restricts the tool further. Press coverage says it halted updates, which we could not verify directly.
• Whether brokerages add controls for sessions that originate from hosted VPS environments.

Accurate as of October 25, 2026, based on GMO's notices as relayed by secondary outlets and on Jiji Press reporting. We could not open GMO's original notices directly; figures may change as the investigation continues.

Related reading

• Japan's September 2026 cyberattacks explained
• October 2026 cyberattacks tracker
• AI agents and Latin American financial targets: what is confirmed
• Bill Gates on AI cyberattacks
• G7 finance tabletop exercises on AI cyberattacks
• Big US banks block AI agents from customer accounts
• Felony Bench: tracking AI agent incidents