Security

Community GPUs are cheap because they run on other people's machines. Here is exactly what that means for your data, what we protect and what we cannot.

What the GPU host can see

Your job runs in a container on a machine owned by an independent host. While it runs, the host can in principle access the container: your code, model, data and outputs. This is true of every GPU marketplace.

• Do not run personal, medical or otherwise confidential data this way.
• Data-center-only mode limits jobs to verified data-center hosts on Vast.ai and to RunPod Secure Cloud. It lowers the risk; it does not remove it — data is still unencrypted while it is being processed.
• Open models and public or synthetic datasets are the right fit.

Your GPU account stays yours

• Jobs run on your own Vast.ai or RunPod account; you pay the platform directly. We never hold your GPU balance.
• We act through an API key you create. Delete it at the platform and our access ends immediately.
• Every machine we rent carries a label with your job name, so you can see it in your platform console.

How we store your keys

• A key you enter in your account is checked, then encrypted with AES-256-GCM and bound to your account, so it cannot be moved to another one. It is never shown again.
• The encryption master key is kept outside the database; backups of the database and of our server configuration are encrypted too.
• Keys never go through the support chat, email or logs. If you paste one into the chat by mistake, we ask you to revoke it.

Checkpoints and storage

• Checkpoints are kept in Cloudflare R2 object storage.
• Each job gets its own temporary storage key that only works for that job's folder and expires after 4 hours (it is renewed while the job runs). A leaked job key cannot read your other jobs.
• Optionally, checkpoints are encrypted with your own key (AES-256-GCM) inside the container before upload, so storage only ever sees ciphertext.
• Jobs started from your account cannot read files or secrets from our server: templates decide what goes into the container.

Our side

• Our server is reachable for administration only over a private network; the operator panel sits behind two independent sign-ins.
• Every action in the panel and in your account is written to an audit log: who, when, from where, on what. Keys and passwords never go into it.
• Account passwords are hashed with scrypt. You get a notification on every sign-in from a new browser, and you can sign in with a one-time code by email or Telegram.
• We never see your card details: card and bank payments go through EasyStaff; crypto payments are checked on the public blockchain.

What we do not claim

• No SOC 2, ISO 27001 or similar certification.
• NomadGPU is a small team; access to production is limited to the operator.
• No system is perfectly secure. If a breach affects your data, we will tell you without undue delay.

Report a problem

Found a vulnerability or something that looks wrong? Write to ops@nomadgpu.com. We reply within one business day and credit reporters who want it.