Compute-to-Data Security Model

Compute-to-data runs third-party code near protected data. Deployments should treat every workload as untrusted until its identity, digest, authorization and policy have been verified.

Required controls

• Verify publisher signatures and immutable workload digests.
• Use isolated, least-privilege runtimes.
• Deny network access unless explicitly required.
• Expose data only through the authorized bucket interface.
• Enforce resource, time and output limits.
• Keep secrets outside images and execution receipts.
• Scan packages and images before making them eligible.
• Sign node execution receipts and hash every retained artifact.
• Record policy denials and abnormal termination.

Disclosure control

Output is a separate authorization boundary. A workload that can query protected records must not automatically be allowed to return those records. Result policies, size limits, schema checks and disclosure review prevent a workload from using its output as an unrestricted data-export channel.

Contracts and workloads

Smart contracts are deterministic consensus logic. Compute workloads are sandboxed data-local jobs that may be resource-intensive or non-deterministic. The blockchain authorizes, coordinates and records them; the local runtime performs the computation.