Technical Deep-Dive
How the Data Sovereignty Model works
A four-layer architecture that lets AI access sensitive data without ever owning it. Here's exactly how.
Architecture
Data Layer
- IPFS/Filecoin encrypted vaults
- Content-addressed storage
- Owner-held encryption keys
Access Layer
- Solidity smart contracts
- Permission management
- Time-bounded access grants
- Purpose-specific scoping
Compute Layer
- Federated learning across vaults
- Zero-knowledge proof generation
- Differential privacy guarantees
Audit Layer
- Immutable access logs on-chain
- Cryptographic attestations
- Regulatory compliance proofs
Step-by-Step Flow
Data owner encrypts and stores sensitive data in an on-chain vault
The owner generates encryption keys and stores the encrypted payload on IPFS. The decryption key is held only by the owner. A smart contract records the vault address and owner identity.
AI system requests access via smart contract
The AI system submits a request specifying: what data fields, for what purpose, for how long. The smart contract checks the request against the owner's permission policy.
Owner approves (or auto-approves via policy)
The owner can approve manually or set policies that auto-approve certain request types. Approval generates a time-bounded, purpose-specific decryption token.
AI processes in-place using federated computation
The AI model runs computation against the decrypted data in a secure enclave. Results are extracted. The raw data is never copied, transferred, or stored by the AI system.
Zero-knowledge proof verifies the result
A ZK proof attests that the computation was performed correctly on the authorized data, without revealing the data itself. The proof is verifiable by any third party.
Immutable audit log records everything
The access request, approval, computation, and result hash are all recorded on-chain. The full history is auditable forever. No deletions, no modifications.
Frequently Asked Questions
Where does my data physically live?
Encrypted on IPFS/Filecoin — decentralized storage. The encryption key is held only by you. Even the storage nodes can't read your data.
Can the AI system make a copy of my data?
No. Computation happens in-place in a secure enclave. The AI gets results, not raw data. Zero-knowledge proofs verify this.
What if I want to revoke access after granting it?
Access tokens are time-bounded and purpose-specific. You can revoke at any time via the smart contract. Revocation is instant and on-chain.
How is this different from traditional encryption?
Traditional encryption protects data at rest. Our model protects data during computation too. The AI never sees raw data — it works with encrypted data and proves its work via ZK proofs.
Is this production-ready?
The Data Sovereignty Model is a research framework. We are building proof-of-concept implementations for each industry vertical. Production deployment timelines vary by use case.