Implicant

I am a bank.

How do I detect fraudulent account applications?

I want

the most
accurate model

I cannot

expose my
customers' data

The Implicant solution

Privacy-preserving AI.

  • The bank encrypts the application.
  • Implicant server or your cloud provider runs the model on the encrypted application and cannot read it.
  • Only the bank can decrypt the answer.

Encrypted. Exactly as accurate. Answered in real time.

Let's run it.

The bank
Private key
Stays at the bank.
Public key
 
internet
Implicant server or your cloud provider
Nothing yet

Made once per bank. The private key, the only one that decrypts, never leaves the bank. The public key lets the server compute, never read.

Pick an application.

The bank Private keyPublic key

Edit any field, then run the inference.

The bank Private keyPublic key

The bank Private keyPublic key
internet
Implicant server or your cloud provider Public key
Nothing has arrived

    Decrypt the result

    The bank Private keyPublic key
    →
    Implicant server or your cloud provider never decrypted the encrypted data

    A conventional AI provider

    what it would see
    27 of 27 fields readable

    Receives the application in the clear.

    Implicant server or your cloud provider

    what it receives
    0 of 27 fields readable

    Receives the same application encrypted, and still computes the answer.