Data Privacy
Our Most Important Mission
NavyAI leverages the transparency of blockchain technology while employing sophisticated cryptographic methods to safeguard user data during input and output phases. This document delineates the encryption protocols established to maintain privacy.
Miners’s asset
Every miner (also referred to as a model host) within the network generates a distinct pair of public and private keys.
Miners disclose their public keys for user access.
Data Encryption by Users
Users create a symmetrical encryption key for the secure encryption of input data destined for machine learning model inference.
This symmetrical key is then encrypted using the public key of each selected miner. Should a user choose N miners, this encryption process is replicated N times with each miner's public key.
The user compiles the encrypted input data, the assortment of encrypted symmetrical keys, and their public key into a submission. This submission is then broadcasted across the network.
Data Handling by Miners
Upon receiving a user's submission, a miner decrypts the symmetrical key using its private key.
With the symmetrical key, the miner decrypts the user's original input data.
The miner processes the model inference task with the decrypted input data.
Once the task is complete, the miner encrypts the result using the user's public key.
This encrypted output is shared on the network, accessible solely by the originating user, who can decrypt it with their private key.
Security Measures
The implementation of symmetrical encryption for input data ensures exclusive access to the original data by the corresponding miner via its private key.
Output data encrypted with the user's public key guarantees that only the user can decrypt and access the model's results.
This layered encryption strategy effectively shields user data from unauthorized viewing at all stages.
The public key infrastructure (PKI) framework confirms that encrypted messages are decipherable solely by the designated recipients, thus securing end-to-end data confidentiality.
Last updated