Fugaku Supercomputer Breach: 430 Keys Stolen

Japan's flagship Fugaku supercomputer remains offline after its operator, the national research institute RIKEN, confirmed that intruders stole 430 credentials belonging to users and accessed the personal details of 4,222 users.
In a second report published on Friday 20 November, RIKEN said the stolen items were credentials that users had stored in their home directories on Fugaku for connecting to other institutions' supercomputers and cloud services, chiefly SSH private keys, along with access keys. Investigators found traces of the keys being collected and moved out of those directories, and RIKEN is treating all of them as leaked.
More than half of the stolen keys had no passphrase protecting them: 233 of the 430, against 197 that did. A private key without a passphrase can be used by anyone who holds it, which is why RIKEN is contacting the owner of each credential individually and asking them to revoke it at the service it belongs to and issue a replacement. Administrators at other computing centres have been told to remove any keys stored on Fugaku from their authorised-keys files, rotate access keys issued to Fugaku users and check logs for logins since 10 November from unfamiliar places.
Personal details of 4,222 users treated as leaked
RIKEN also found traces of the intruder accessing the registration information of 4,222 Fugaku users, and it now treats that information as leaked too. Besides names, email addresses and user IDs, the records include the address, affiliation and phone number each user gave when registering, and a record of which type of document they showed to verify their identity. RIKEN said it has found no confirmation that copies of the identity documents themselves leaked.
The institute said it has seen no other misuse so far beyond the attempts to connect to other systems using some of the stolen keys, which it had already reported. It said it has no concrete evidence that research results leaked. It has not named who was behind the attack, or said which other institutions' systems the stolen keys open.
Intruders entered through a public dashboard
Between 10 and 13 November, an outside attacker used a web dashboard that showed Fugaku's operating status and was exposed to the internet as a foothold into the internal operations network. RIKEN said a software vulnerability was exploited but has not named it. Inside the network, the attacker operated several servers and planted malicious programs on some of them. Maintenance of the affected system is outsourced to an external contractor.
Suspicious traffic was detected on 12 November. RIKEN confirmed unauthorised access at 16:33 Japan time on 13 November, isolated the system used for the intrusion that evening, and stopped Fugaku and cut off outside access at 04:30 on 14 November. All running and queued compute jobs were halted, administrative keys were replaced, and on 15 November users were emailed and every user's login public key was invalidated.
The first notice, published on 16 November, had put the number of users who may be affected at up to 4,597, including up to 375 users of Rikyu, a newer AI-for-science supercomputer that RIKEN runs. The second report gives 4,222 affected Fugaku users and no new Rikyu figure.
No date set for Fugaku's return
RIKEN says Fugaku will return only after it separates its public services from the management network and rebuilds servers, and it has not set a date for the restart. It is setting up an investigation committee that includes outside experts and drawing up a reform plan. It has briefed the Saitama and Hyogo prefectural police and reported the incident to Japan's Personal Information Protection Commission, the science ministry and the organisations behind HPCI, the national research computing network that links Fugaku with university supercomputers.
RIKEN also warned users about emails pretending to come from RIKEN or Fugaku, and said it never asks for passwords or similar information by email. Fugaku, based in Kobe, began full-scale operation in 2021 and has been used for a wide range of research, including simulations for COVID-19 countermeasures, weather forecasting and drug discovery.