Cyber & AI intelligence
Briefs indexed3924
Issues37
Published Mondays07:30 CT
▣ Breach RIKEN-FUGAKU-SUPER 2026-11-18

RIKEN Fugaku: Public Status Dashboard Breach Exposes Up to 4,597 Users and Stored SSH Keys

"Japan's national research institute RIKEN disclosed on November 16 that an intruder broke into the operational network of Fugaku, the country's flagship supercomputer. The attacker used Fugaku's public "operating…"

Japan's national research institute RIKEN disclosed on November 16 that an intruder broke into the operational network of Fugaku, the country's flagship supercomputer. The attacker used Fugaku's public "operating status" dashboard as the way in. Between November 10 and 13, they reached several internal servers, installed malicious programs and took, or may have taken, user data. The exposed data includes SSH private keys and access keys that researchers had stored for other supercomputers and services. RIKEN's own notice puts the maximum exposure at 4,597 people: up to 4,222 registered Fugaku users and up to 375 users of RIKYU, RIKEN's separate AI-for-science supercomputer. RIKEN has taken Fugaku offline and says it has not yet worked out exactly how the intruder broke in. No one has claimed the attack and RIKEN has not attributed it.

What Happened

RIKEN published its notice in Japanese on November 16. An English version followed on November 17. It lays out the timeline below. RIKEN had outsourced the maintenance and management of the affected operational system to an outside contractor.

Media accounts differ slightly on the timeline. Tom's Hardware says the investigation confirmed unauthorized access "by November 14". RIKEN's own record puts confirmation at 4:33 p.m. on November 13. The Register reports that someone tried to log on to Fugaku itself on November 13, which prompted RIKEN to take the machine offline. The Japan Times reports that Chief Cabinet Secretary Minoru Kihara said on November 17 that RIKEN and MEXT were investigating the cause and scope of the breach. He called Fugaku vital to Japan's research and development infrastructure.

What Was Taken

RIKEN describes the data as "taken or may have been taken". Its figures are upper limits, not confirmed counts of stolen records. The notice lists two groups of data.

1. User authentication system records (up to 4,597 people) - Fugaku users (up to 4,222): names, email addresses and user IDs - RIKYU users (up to 375): names and email addresses

2. Files in Fugaku users' home directories - Credentials that users had stored for connecting to supercomputers and services outside RIKEN, including SSH private keys and access keys - Research data files

Coverage varies in how firmly it states the loss. BleepingComputer lists the 4,222 and 375 figures as compromised data. The Japan Times says the RIKYU users' information "was accessed". RIKEN's own wording is more careful. It says these are the maximum numbers of users who "may have been affected". RIKEN also says it has found no specific evidence that research results have leaked, and it is still checking whether the affected files contain personal data about third parties.

The most serious part is the credentials. RIKEN confirmed that the intruder used some of the stolen credentials to try to connect to other systems. It is working with the organisations concerned to find out whether any of those attempts succeeded. The Register reports that this includes the operators of other supercomputers that RIKEN does not run. As of RIKEN's notice, no follow-on compromise has been confirmed or ruled out.

Why It Matters

This was a credential theft, not just a data leak. The names and email addresses matter less than the SSH keys and access keys found in researchers' home directories. HPC users commonly keep keys on shared systems to move data and jobs between machines. A single breach at one centre can therefore expose accounts at other centres, cloud services and partner institutions. Those organisations did not suffer the original breach but now carry the risk.

A small, unglamorous system was the way in. A public page showing machine status was enough to reach the internal operational network. Status pages, monitoring dashboards and similar tools are often treated as low priority, are often run by contractors, and are often connected to the networks they report on. In this case that connection gave the attacker access to user storage.

The operational cost is large. Fugaku is a 537-petaflop system, according to BleepingComputer. It is used for drug discovery, severe-weather forecasting and simulations of airborne virus transmission. It led the Top500 list from 2020 to 2022 and ranks ninth on the list released this week, according to The Register. It is now offline with no restart date. RIKEN says operations will resume only once it is safe. The Register notes that an attacker able to run workloads on a machine of this scale could do considerable damage, but no one has reported that this happened.

Responsibility was shared with a contractor. RIKEN outsourced the maintenance of the affected system. Detection and investigation depended on coordination with that contractor. Any organisation that outsources the running of its infrastructure should review where this kind of arrangement leaves gaps in visibility.

The Attack Technique

What is confirmed is limited, and RIKEN has said so directly.

RIKEN has not named a threat actor, and no group has claimed the intrusion. According to BleepingComputer, RIKEN's recovery plan covers patching the exploited vulnerability, rebuilding compromised servers, reviewing the network configuration and adding AI-driven security monitoring.

What Organizations Should Do

  1. Rotate any credential that was ever stored on Fugaku. RIKEN has asked users to revoke credentials kept on Fugaku at each service they connect to. Partner HPC centres and cloud providers should go further. They should find accounts whose keys could have been stored on Fugaku, check logins from November 10 onward for unfamiliar sources, and force a key change where there is any doubt.
  2. Separate status and monitoring pages from production networks. Public dashboards should get their data from a one-way feed or a separate copy, not run inside the network they display. Treat every internet-facing management or status page as an entry point and patch it on that basis.
  3. Stop users storing reusable private keys on shared systems. Move users to short-lived SSH certificates, agent forwarding with confirmation prompts, or hardware-backed keys. Regularly scan home directories for unencrypted private keys and cloud access keys, and alert on them.
  4. Watch internal traffic as closely as the perimeter. RIKEN caught this through suspicious traffic on November 12, up to two days after the intrusion started. Alerts for unusual internal connections from front-end servers into authentication and storage systems can shorten that gap.
  5. Set contractor responsibilities in writing. Where contractors run operational systems, contracts should state who handles log access, detection, escalation deadlines and incident response. That way the asset owner can act within hours, not days.
  6. Rehearse a full shutdown. RIKEN's steps (isolate the entry system, replace administrative keys, disable all user login keys, stop jobs, notify regulators and partners) are a sound response. HPC operators should have that procedure documented and tested before they need it.

Sources: Hackers Breach Japan's Fugaku Supercomputer, Exposing Data of Up to... | Japan takes Fugaku supercomputer offlline after cyberattack | Japan's Fugaku supercomputer has decidedly un-super security | Notice regarding a possible personal information leak affecting Fug... | Japan’s most powerful supercomputer hit by unauthorized access - Th... | Japan’s most powerful supercomputer may have been hacked, over 4,00... | HPCwire - Since 1987 – Covering the Fastest Computers in the World... | スーパーコンピュータ「富岳」および「理究(りきゅう)」 への不正アクセスによる個人情報漏えいの可能性について 理化学研究所