Onamae Desktop Cloud
Twenty six thousand servers, and nobody at the hosting company knows what is on any of them.
What happened?
Onamae Desktop Cloud, a Japanese cloud hosting provider, confirmed on 26 October 2026 that intruders had compromised its infrastructure and installed malware on customer servers. More than 26,000 servers were potentially affected.
The distinction that matters is where the malware landed. This was not a breach of the provider's own customer records. The compromise reached down into the tenant environments, which is to say into the machines that customers rent and operate as their own.
What data was actually inside?
Unknown, and unknowable from the provider's side. That is the defining characteristic of this incident rather than a gap in the reporting.
A hosting provider sells capacity. What a tenant puts on a rented server is the tenant's business, and the provider has neither visibility into it nor, under most agreements, any right to look. Twenty six thousand servers in a Japanese hosting estate will include accounting systems, customer databases, medical practice software, law firm document stores, ecommerce back ends and a long tail of machines whose owners have forgotten what is running on them.
Who gets hurt and how?
Tenants, and then the people whose data those tenants hold. The harm is distributed across 26,000 separate blast radii that have nothing in common except a hypervisor.
Malware installed on a server is a more serious condition than data copied from a database. It implies persistence and the capability to continue acting, which means every tenant must assume their workload was both read and modified until they prove otherwise. For a small business with no security function, which describes a large share of any shared hosting customer base, that determination is beyond what they can perform.
What did they think they were doing right?
The shared responsibility model, applied correctly. The provider secures the infrastructure, the tenant secures the workload, and the line between them is written into the contract. That division is the industry standard and it is not wrong.
What the model assumes is that the boundary holds in both directions. The tenant accepts that the provider cannot protect what is inside the instance. The implicit trade is that a failure at the infrastructure layer stays at the infrastructure layer. Malware reaching customer servers from the provider side is exactly the event the division of responsibility does not account for.
What did they not know about their own data?
The provider does not know what is on the servers, by design. The tenants frequently do not know either, which is the part that converts an infrastructure incident into an unresolvable one.
A server rented in 2019 for one application, repurposed twice, carrying a database dump somebody made before a migration and never deleted, is an ordinary object in any hosting estate. Its owner can describe what the machine is for. Far fewer can describe what is on it.
So the question of what was exposed has no authority who can answer it. The provider cannot see inside. The tenant can see inside but has not looked. The Supabase entry in this series documents the same structure, with the same silence at the point where an answer should be.
If you use cloud storage, do you know what sensitive data lives in your buckets and blobs? Or would you find out the same way they did?
Not sure where to start? How to find sensitive data across your estate
What does attribution look like the morning after?
No threat actor has been named. The provider's notification obligation runs to its direct customers and is comparatively simple: it can enumerate its tenants.
Every obligation past that point belongs to someone else. Each of 26,000 tenants must independently determine whether personal data was on the affected machine, under Japan's Act on the Protection of Personal Information or whichever regime covers their own users. The provider cannot perform that assessment for them, and most tenants lack the means to perform it at all. The notification chain breaks at the point where visibility was contractually transferred.
What would have changed the outcome?
On the tenant side, knowing what is on the machine. The shared responsibility model assigns the tenant the duty to secure the workload, and that duty is impossible to discharge without an inventory of what the workload contains.
The useful exercise is unglamorous and available today. For each rented instance, what sensitive data is on it, why, and when was that last verified. Organisations that can answer for every instance treat an event like this as scoped work. Organisations that cannot are waiting for somebody to tell them something nobody is in a position to say.
Onamae Desktop Cloud found out the hard way.
Your team could spend the next 6 months rebuilding systems, notifying customers, and answering legal questions. Or you could spend 24 hours finding out what's actually at risk.
Want to understand the process first? Read our guide to sensitive data discovery .