Secure AI operations
Assignee01 Quantum Inc.
InventorsAndrew Cheung, Alexey Shpurov
Life of this patent
Estimating the term…
Abstract
An artificial intelligence model marketplace system secures both models and user data by managing computation-preserving-cryptography (CPC) encryption of data and models with public and private CPC user key pairs. Models can be encrypted by the marketplace system, model vendors, or third parties. e.g., in response to user requests to apply a model to their data to obtain a model prediction. Similarly, execution of the CPC encrypted models on CPC encrypted user data can be performed by the marketplace system, model vendors, or third parties. Selection of third parties for encryption and/or model execution work can be organized via staking mechanisms to create a secure, distributed, competitive marketplace for computational services.
Claims8
A system comprising a computer server configured to:
- solicit, from a plurality of model vendors, descriptions artificial intelligence (AI) models provided by the model vendors;
- receive, from two or more of the model vendors, descriptions of available AI models;
- provide, to one or more model users, the descriptions of the available AI models;
- receive, from a model user over a first post-quantum cryptographically (PQC) secure session, a request and user data, wherein:
- the request is a request for a prediction to be generated by a selected AI model,
- the selected AI model is selected by the model user from the available AI models,
- the user data is encrypted using a computation-preserving-cryptography (CPC) public key of the model user, and
- the CPC public key is associated with a CPC private key of the model user;
- responsive to receiving the request, obtain the CPC public key of the model user from a key exchange server (KES);
- obtain, over a second PQC secure session from a selected model vendor providing the selected AI model, an encrypted version of the selected AI model, wherein the encrypted version of the selected AI model is encrypted, using the CPC public key of the model user, by the selected model vendor by the selected AI model;
- generate, using the encrypted version of the selected AI model on the encrypted user data, the prediction, such that the prediction is inherently encrypted using the public CPC key of the model user;
- and
- provide, to the model user, the encrypted prediction.
A system comprising a computer server configured to:
- solicit, from a plurality of model vendors, descriptions artificial intelligence (AI) models provided by the model vendors;
- receive, from two or more of the model vendors, descriptions of available AI models;
- provide, to one or more model users, the descriptions of the available AI models;
- receive, from a model user over a first post-quantum cryptographically (PQC) secure session, a request and user data, wherein:
- the request is a request for a prediction to be generated by a selected AI model,
- the selected AI model is selected by the model user from the available AI models,
- the user data is encrypted using a computation-preserving-cryptography (CPC) public key of the model user, and
- the CPC public key is associated with a CPC private key of the model user;
- responsive to receiving the request, obtain the CPC public key of the model user from a key exchange server (KES);
- obtain, over a second PQC secure session from a selected model vendor providing the selected AI model, an unencrypted version of the selected AI model;
- encrypt, using the CPC public key of the model user, the unencrypted version of the selected AI model to create an encrypted version of the selected AI model;
- generate, using the encrypted version of the selected AI model on the encrypted user data, the prediction, such that the prediction is inherently encrypted using the public CPC key of the model user;
- and
- provide, to the model user, the encrypted prediction.
Dependencies are inferred from the claim text.