.png){kind=small}
Our code can be executed outside of your network, but your raw data stays inside. We use either 100% software techniques, or, cheap, hardware security modules (HSM) which we deliver to our clients. Our solution can be used for basic use-cases, such as confidential medical messaging
we give you the keys to keep ownership of your data, be it stored in the cloud or in your servers. While at the same time enabling your partners to process it in a predefined way, without ever seeing it. E.g. outsourced/third party applications, or external models trained/applied on your data
We enable our clients to train a model collaboratively, which is encrypted so that predictions are computable only if every client is ok. We enable our clients to quickly access joint predictions of proprietary models, by protecting these models with differential privacy (trustless PATE framework)
We provide our clients with a platform for automatizing all these computations between them, where they can buy and sell these computations autonomously
We secure the models obtained by keeping the model under an encrypted form. So that predictions are computable only if every partner is ok. We help you specify the acceptable requests to the model to avoid reverse engineering. In the case of many data owners, we provide federated learning without the need of a trusted aggregator
We enable you to outsource computations and enable queries from your clients in the cloud where your datais stored, without exposing it to the cloud provider. Our technical solution enables to encrypt/decrypt remotely in a virtual machine of the cloud, with sovereign keys. As an option, the metadata can be left in clear to allow NoSQLqueries. We provide one key-per-file at a flat cost, so you can manage fine-grained access control by design on who can process each file
We protect costly models with watermarks and auditable Zero-knowledge proof of copy
Differentially private and secure aggregation of private models predictions, without disclosing your data