FHE fog-of-war experiments

Build the repo:

cargo build --bin fhe-zama --release

Test out distance calculations and speed of the tfhe-rs library

./target/release/fhe-zama basic

Threshold FHE example with mock MPC network

Mock MPC(t=1, n=3) network and FHE example.

./target/release/fhe-zama mpc -t 1 -n 3

Which outputs

Alice:
        Encrypting Alice's Position { x: 2, y: 2 } with FHE client_key
        Serializing and encrypting position with MPC public_key...
        ==> Sending to MPC_Network

Bob:
        Encrypting Bob's Position { x: ?, y: ? } with FHE client_key
        Serializing and encrypting position with MPC public_key...
        ==> Sending to MPC_Network

MPC_Network:
        Fetching MPC shares and decrypting for FHE ciphertexts...
        Running FHE operations on Position ciphertexts...
        Alice's fog-of-war view range: 11
        should_reveal_bob?: true

        Bob is within Alice's FOW view range, decrypting Bob's position...
        Encrypting response and sending to Alice...

Alice:
        Alice received and decrypted Bob's Position { x: 4, y: 4 }

Time elapsed: 12.706566s

In the above example should_reveal_bob is a FHE ciphertext the MPC network decrypts and sees is true. Then conditional on should_reveal_bob = true, the MPC network decrypts Bob's position and re-encrypts it for Alice using Diffie-Hellman.

• This is sup-optimal as the MPC network is trusted with decrypting Bob's position.
• How do you know the MPC network decrypted just the should_reveal_bob boolean ciphertext, and not anything else?
• We would need a way to Zk-prove that the MPC network decrypted Bob's position if and only if should_reveal_bob = true

Key issue: conditional decryption

Need a way to conditionally reveal encrypted state to some users, without the MPC nodes seeing plaintext.

• Decrypt Bob's position just for Alice to see, conditional on Bob being "close" to Alice.

Perhaps private-set-intersection (PSI) can help: PSI:

• PSI being used for confidential "coincidence of wants" in social apps:
  • tinder
  • job matching
  • auctions
  • liking/upvoting risky content / political views