Abstract
This disclosure describes a custody system for data and AI models designed so that a ransomware attack can stop a node but cannot destroy what it holds or delay its recovery indefinitely. Six measures work together. First, the restore vault pulls copies from the primary system; the primary has no path to write to, alter or delete anything in it, and retention periods are time-locked in the vault's own firmware so that no one, including administrators, can shorten them. Second, an independent safety plane, on its own power, processor and channel, watches for the signs of encryption by ransomware, such as sudden mass rewrites, a jump in the randomness of written data, and changes to canary files that no legitimate process touches; on detection it isolates the node by itself and preserves the latest good restore point. Third, recovery takes place only in a clean room: on freshly provisioned or re-imaged hardware that must first pass attestation, never in the environment that was attacked. Fourth, where a depositor's terms permit, encrypted replicas held at custody nodes in other locations serve as further recovery sources; because they remain encrypted under the depositor's key, those peers cannot read them. Fifth, administration is isolated, using hardware keys, a separate management network and a two-person rule. Sixth, recovery drills are a condition of certification, and each node's recovery readiness is published in a public registry. Responsibility for creating, protecting and testing restore points stays with the operator by rule and cannot be shifted to depositors by contract.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Data Bank™: Ransomware-Resilient Data Custody with a Pull-Only, Time-Locked Restore Vault, Independent Detection, Clean-Room Recovery and Encrypted Peer Restore", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12169