Abstract
This disclosure describes the identifier on which a governed data custody architecture keys all of its functions. Every unit of data deposited into custody, whether a file, a data set, a record collection or a time segment of a continuous sensor stream, receives a data identifier when it is deposited. The identifier is bound by a signed record to the hash of the content, the version, the depositor, the depositor's terms and the permitted region, and contains no personal information itself. The identifier stays the same when the deposit moves between custodians, is restored from a copy, or is re-encoded without change of content; a change of content creates a new version with its own identifier linked to the previous one. Data derived from deposits, such as cleaned, labelled, sampled, aggregated or synthetic data, receives its own identifier linked to every source identifier and to the transformation used. Each time a certified AI system trains, is evaluated or runs inference on a deposit, the link between the data identifier and the model's lifelong identifier is recorded, so that the identifier graph runs from data to model to every system running that model. Restore points, key shares, residency rules, usage credits, protection status and transfer records all refer to deposits by this identifier. Anyone may resolve an identifier to its live status, such as active, superseded, withdrawn, flagged as poisoned or under transfer, without learning its content or its depositor unless the depositor allows. When a deposit is withdrawn or found poisoned, the identifier graph shows every derivative and every model affected, so that each can be retrained, restricted or recalled. The identifier is also the anchor of the data's security: certified systems refuse data without a valid identifier, the content hash is checked on every read and restore, keys are released only while the identifier's status allows, a poisoning flag quarantines everything linked to it, access is monitored per identifier, and leaked data can be traced back to its deposit.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Data ID™: A Persistent, Verifiable Identifier for Every Data Deposit, Version, Segment and Derivative, Binding Content Hash, Depositor, Terms and Region, Linked to Every Model Trained on It, and Resolvable by Anyone to Live Status but Never to Content", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12356