The use of Fully-Homomorphic Encryption (FHE) allows a user to encrypt the data, send it secured to the AI model and to let the AI model compute over encrypted data. Nobody between the user and the model, and not even the model provider can spy on sensitive information from the user. The result can only be read by the user who has the encryption keys.
Zero-Knowledge Proofs (ZKPs) then provide additional security by providing a digital receipt that the model executed the request you wanted to execute. ZKPs provide integrity validation on the model execution.
Dropnir serves as a thin layer between the application and the model.
As a user, you don’t notice any difference in behavior from talking to the model or to Dropnir.
The software is easy to install (as a container) and serves as a thin API layer.
The model performs its magic, whatever it is meant to do, but it does it over encrypted data!
The model then sends an answer back to Dropnir, again fully encrypted. The owner of the model was never able to view the request, the data, or the answer.
Dropnir decrypts the result using the private key from the customer, and provides a clear-text readable answer to the customer’s application.
Wodan’s SaaS platform is used for metrics collection on how much data and requests are transmitted through Dropnir, and provides easy chargeback.
