bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 45쪽 중 2쪽
n-making applications. This establishes a
robust framework to safeguard decision-making processes against both external and internal threats,
not only removing the need for trust among participants, but also preserving the confidentiality and
integrity essential for democratic decision-making environments.
Structure of CRISP Secret Ballots
This section details the components built atop Interfold’s general-purpose core in order to enable
modern, coercion-resistant ballots. This includes smart contracts that mediate Requester and Data
Provider interactions with Interfold, along with a vote-tallying computation to be run via
Interfold.
The CRISP implementation includes a Zodiac
18
-compatible module that can be used to control any
contract account that conforms to the Zodiac IAvatar interface; a Safe
19
, for example.
Setup
To enable a compatible account for control by CRISP secret ballots, one must deploy a CRISP module
and activate it on the account to be controlled. This process involves specifying the following:
The account to be controlled by the module.
The ID of the designated vote-tallying computation.
The designated execution environment, along with any additional parameters.
The required number and threshold of Ciphernodes.
The duration for each poll.
The address of a voter registry contract.
The specific steps for enabling a module on a smart contract account may differ between
implementations and are not detailed in this document. Once the CRISP module is enabled on the
account, it can be utilised to create polls that can ultimately trigger the account to make any
arbitrary call.
Poll Creation
Anyone can propose a transaction to be executed by a smart contract account with a CRISP module
enabled. To do so, one must call a function on the CRISP module and provide the following: the hash
of the proposal description, the hash of the transaction payload, and the bond for the computation
required by the CRISP network.
This will register the proposal in the CRISP module and request the computation from the CRISP
network using the threshold and duration parameters defined in the CRISP module’s setup.
Voter Registration
In each poll, every voter must have a cryptographic keypair to cast their vote, which must be
registered in a Voter Registry smart contract. This contract is responsible for:
Specifying the necessary proof for validating messages in the FHE computation.
Enforcing any voter eligibility criteria mandated by the poll.
Determining the vote weight for each registered voter.
For example, a Voter Registry may require that voters hold a minimum amount of a token or be
provably a member on a predefined list of eligible voters. Importantly, voting keys are distinct
from the keypairs a voter might use to otherwise interact onchain; voting keys are single-use keys
specific to the current poll, disposable once the poll concludes, and should not be reused for other
polls or contexts.
Voting
Votes are submitted directly to the CRISP smart contract as ciphertext encrypted to the shared key
provided by the selected Ciphernodes, along with a zero-knowledge proof that the encrypted message
represents a valid vote format was signed by a registered voter and cannot be reused in any other
proposal. This setup allows any account to submit a vote message on behalf of any user.
Voters have the option to change their vote at any time prior to the Input Deadline by submitting an
encrypted message to the CRISP smart contract. Only the latest message will be counted in the
tallied results.
Key Switching
Voters have the option to change their voting keys at any time prior to the Input Deadline by
submitting a correctly formatted key change message to the CRISP smart contract, encrypted to the
private key provided by the selected Ciphernodes. Only messages signed by a voter’s most recent
valid key will be counted in the tallied results.
Computing and Publishing Results
Once the Input Deadline has passed, the selected Compute Provider will compute the tally ciphertext,
and the selected Ciphernodes will both decrypt the output and post the hash of plaintext results
onchain.
Triggering Onchain Actions
Once the selected Ciphernodes post the hash of the plaintext results onchain, anyone can call a
function to execute the attached transaction payloads. This execution requires providing proof of
the results supplied by the selected Ciphernodes.
Mitigation of Common Attacks
Our voting system built on CRISP is designed to be resilient against a host of practical and
theoretical attacks. This section details several of these attack vectors, along with the
corresponding mitigations in our voting implementation.
In each scenario, Alice and Bob are both registered voters. Alice and Bob are both outspoken
supporters of the Banana Party.
Chuck runs the web service through which voters submit their votes in polls and supports the Durian
Party. Other characters may also be introduced.
Censorship
Knowing that Alice and Bob will each likely vote for the Banana Party, Chuck chooses to ignore
Alice’s and Bob’s votes when they are submitted to his web service via the voting application,
neglecting to post the votes on chain.
As Chuck was unable to provide a valid transaction receipt to Alice and Bob, showing that their
votes had been submitted onchain, Alice and Bob can choose to post their votes directly to the CRISP
contract without permission or censorship from any intermediaries, like Chuck. By submitting their
votes directly or via an alternate relaying service, Alice and Bob can circumvent Chuck’s attempt to
censor their votes.
Chuck could also attempt to mount a similar attack attempting to deny voter registration for any
given poll, but the mitigation would be similar.
Receipt Sharing
Knowing that Alice and Bob will each likely vote for the Banana Party, Chuck decides to offer Alice
a bribe to vote for the Durian Party, rather than her original preference.
Alice’s optimal behavior is to accept the bribe from Chuck, vote in the same way they would have
without the bribe, and simply supply Chuck with a fake proof that is indistinguishable from a
legitimate vote. To do this, Alice could vote for the Durian party, then submit a keychange message
to change their voting key, invalidating their previous vote, and then finally cast a vote for the
Banana party with the new key. Alice can share the receipt of the first vote with Chuck. However,
Chuck has no way of guaranteeing whether the key used to cast the vote was Alice’s valid voting key.
So Chuck must simply take Alice’s word for it.
Proxy Voting
Knowing that Alice and Bob will each likely vote for the Banana Party, Chuck decides to offer Alice
a bribe to vote for the Durian Party, rather than her original preference. Chuck offers to pay the
bribe to Alice on the condition that Chuck is granted permission to cast Alice’s vote. To comply,
Alice must submit a key change message switching her voting key to one supplied by Chuck.
As with the case of receipt sharing, Alice’s optimal behaviour is to accept Chuck’s bribe and to
share the receipt of a corresponding key change message with Chuck, after having already switched
voting keys to another key unknown to Chuck. Unfortunately for Chuck, there is no way to ensure that
Alice has not previously registered or switched to a different voting key, and no way to determine
if the message to change keys to Chuck’s voting key is in fact invalid.
Forced Abstention
Knowing that Alice and Bob will each likely vote for the Banana Party, Chuck decides to force both
Alice and Bob to abstain from voting (either by bribery or more coercive forms of collusion). To
comply with Chuck’s demand, Alice and Bob must not be caught registering for or casting a vote in
the poll.
Alice and Bob’s optimal behaviour is to vote as normal, while taking care to not leave any
identifiable traces. When registering and submitting votes onchain, Alice and Bob can submit
messages through a relayer other than the one controlled by Chuck or from a fresh address which
cannot be linked to their identity. Registration involves providing a proof that the voter is on the
registry, but does not require the voter to identify themselves in plaintext or submit the
registration message from a specific account. Similarly, when casting a vote, nothing identifiable
is published in plaintext, and no ciphertext aside from the result will ever be decrypted by the
Ciphernodes.
Limitations
Corrupt Registry
Like any other voting implementation, CRISP is dependent on a functioning voter registry. If the
voter registration process is compromised, allowing an attacker to either deny registration or take
control of a voter’s account before they select their voter keys, then the attacker can successfully
corrupt the system.
The latter type of attack could be mitigated in a variety of ways, depending on the scope and trust
assumptions appropriate for a given poll. However, a general rule of thumb is to ensure enough other
value is associated with each voter's account that they would be unwilling to share the credentials
with a third party. At the very least, this makes such attacks more costly and less scalable.
Comparison to MACI
CRISP’s design is heavily inspired by MACI
20
, a protocol initially proposed by Vitalik
Buterin
21
for coercion-resistant voting leveraging zero-knowledge proofs (ZKPs). CRISP differs
from MACI primarily by employing fully homomorphic encryption (FHE) and threshold cryptography. This
approach allows CRISP to establish an arbitrarily large network of nodes for trust distribution,
contrasting with MACI's reliance on a single trusted coordinator.
While MACI relies on an honest coordinator assumption for privacy, meaning the coordinator has
unrestricted access to all of the inputs and intermediate states and is trusted not to divulge them,
CRISP provides strong economic guarantees around privacy, as Ciphernodes are subject to slashing if
anyone can prove that they have attempted to decrypt any input or intermediate ciphertext in a
computation.
Exploration of Potential Use Cases
The following list is an exploration of potential use cases for Interfold.
Sealed-Bid Auctions
An auction type where all participants submit bids without knowing the bids of the other
participants. The highest bid wins, but the amount of the winning bid is only revealed after all
bids are submitted.
Benefits: ensures fairness and confidentiality, prevents coercion and bid manipulation
Receipt-Free Delegation
A voting system where voters can delegate their votes without revealing their choice(s).
Benefits: protects voter privacy and encourages people to participate in democratic processes
Research on Private Medical Data
A method of analysing sensitive medical records and health data from multiple sources for research
purposes.
Benefits: advances medical research securely, compliant with regulations, and preserves patient
privacy
Training AI Models on Multiple Private Datasets
A system that allows training AI models using sensitive datasets from multiple sources without
exposing the raw data.
Benefits: enhances AI capabilities and ensures privacy and regulatory compliance
Collaborative Data Analysis
A method of jointly analysing datasets from multiple organisations without sharing proprietary,
sensitive, or private data.
Benefits: enables new collaborative possibilities between private/public institutions, maintains
competitive advantage, preserves data privacy
Privacy-Preserving Surveys
A way to conduct surveys that ensures the anonymity of respondents and the confidentiality of their
responses.
Benefits: increases participation by guaranteeing privacy, encourage honesty without fear of
retaliation
Collective Climate Modelling
A method for organisations, NGOs, and governments to collaborate on climate models, especially in
cases where sensitive internal datasets — such as proprietary models or competitive data — need to
be kept private for whatever reason.
Benefits: enables cross-border collaborations, protects proprietary models and competitive
information, maintains integrity and accuracy of climate modelling.
Score Aggregation Systems
A system for aggregating scores or ratings in contexts like credit scores, product reviews, Michelin
stars, or web-of-trust networks, without relying on a central operator or exposing individual
inputs.
Benefits: ensures privacy, prevents coercion and manipulation, and increases the legitimacy of
outcomes in scoring and rating systems.
Federated Learning Across Multiple Servers
A method of training AI models across decentralised devices or servers with local data samples
ensures the computation process is encrypted and decentralised.
Benefits: leads to more comprehensive models, addresses remaining privacy/security concerns for
Federated Learning, maintains data privacy
Contributors
A special thank you to the following people for their early contributions to and reviews of the
Interfold whitepaper:
Alex Espinosa, Anthony Leutenegger, Casey Gardiner, Daniel Ospina, Disruption
Joe, Gabriel Shapiro, Giacomo Corrias, Koh Wei Jie, Mike Chan, Pandy Marino, Vitalik Buterin, Yuet
Loo Wong
Footnotes
RISC Zero is a general-purpose, zero-knowledge virtual machine.
Learn more
(opens in a new tab)
↩
SP1 is Succinct's zero-knowledge virtual machine.
Learn more
(opens in a new tab)
↩
Jolt is an extensible zero-knowledge virtual machine by a16z.
Learn more
(opens in a new tab)
↩
Nexus is a modular, extensible, open-source, highly-parallelized, prover-optimized zkVM.
Learn more
(opens in a new tab)
↩
Reality.eth is a crowd-sourced on-chain smart contract oracle system by Reality Keys.
Learn more
(opens in a new tab)
↩
Kleros Oracle is a product that combines Kleros's dispute resolution system with Reality.eth's
cryptoeconomic mechanism for verifying real world events on-chain.
Learn more
(opens in a new tab)
↩
API3 creates decentralized APIs for web3.
Learn more
(opens in a new tab)
↩
Staking deposit amounts are set by Interfold governance.
↩
Registration delay period is set by Interfold governance.
↩
Decommission delay period is set by Interfold governance.
↩
Ciphernode cliff and vesting schedule is set by Interfold governance.
↩
Inactive Node penalty amounts are set by Interfold governance.
↩
Time required to provide the ciphertext output is constrained by the complexity of the
computation, the way the computation scales, and the number of inputs.
↩
Failure to Report percentage is set by Interfold governance.
↩
Ciphernodes must provide the decrypted output within a fixed window of time after the output
ciphertext has been posted by the selected Coordinator Nodes.
↩
The same inputs must always produce the same outputs.
↩
The intermediate slashing penalty, along with its burn ratio, is set by Interfold governance.
↩
Zodiac is a standard for modular and composable smart contract accounts, specified in
ERC-5005
(opens in a new tab)
.
A library and implementation can be found on GitHub
(opens in a new tab)
.
↩
Safe is the most widely used smart contract account implementation,
the code for which can be found on GitHub
(opens in a new tab)
.
↩
Minimal Anti-Collusion Infrastructure
(opens in a new tab)
↩
https://ethresear.ch/t/minimal-anti-collusion-infrastructure/5413
(opens in a new tab)
↩
FOLD Transferability
## /
Encrypted Execution Environments
Shared results from
private inputs
private inputs
private inputs
.
The Interfold is an open-source protocol for confidential coordination. Independent parties encrypt their inputs, a committee of ciphernodes runs the computation, and only the verified output is decrypted.
Start building
Run a ciphernode
Learn the protocol
Key generation.
The committee runs DKG and makes one shared public key.
Choose a path
What do you want to do?
Each path starts with a short overview and continues to step-by-step procedures.
Learn
What an E3 is, how the architecture fits together, and the cryptography that protects the
inputs.
Concepts
Build
Install the CLI, make a project from the template, and write an E3 program with its contracts
and client.
Developers
Run a ciphernode
Set up a node, bond FOLD, get tickets, and serve on committees. Upgrade, migrate, and exit
safely.
Operators
Reference
CLI commands, node configuration, contract addresses, and a glossary of protocol terms.
Lookup
How it works
One E3, from request to result
The Interfold contract records each E3 as it moves through its stages. Ciphernodes act at each
stage and publish proofs on-chain.
Requested
A requester pays the fee and asks for an E3.
CommitteeFinalized
Sortition selects a committee of ciphernodes.
KeyPublished
The committee runs DKG and publishes one public key.
CiphertextReady
The program runs over the encrypted inputs. The encrypted output goes on-chain.
Complete
T + 1 key shares decrypt the output, and a proof confirms it.
Committees
Security comes from the committee
No single ciphernode can decrypt. The protocol supports three committee sizes. Decryption needs
T + 1 key shares.
3
minimum
N = 3, T = 1. The smallest committee.
9
micro
N = 9, T = 4.
19
small
N = 19, T = 9. Higher assurance.
Quick start
A working E3 in four commands
The template starts a local chain, local ciphernodes, the servers, and a web client. See the
Quick Start
for prerequisites.
~/code
Copy
# Install interfoldup, then the interfold CLI
$
curl -fsSL https://raw.githubusercontent.com/theinterfold/interfold/main/install | bash
$
interfoldup install
# Make a project from the default template and start everything
$
interfold init my-first-e3
$
cd my-first-e3 && pnpm dev:all
# The client is at http://localhost:3000
Go deeper
Inside the protocol
E3 lifecycle
Each stage, the events it emits, and what happens when a stage times out.
Cryptography
Threshold BFV, zero-knowledge proofs, and the circuits that tie them together.
Sortition
How the protocol selects a committee from ticket balances.
CRISP
A complete E3 application for private voting.
## /learn
Learn
Overview
What is the Interfold?
The Interfold is an open-source protocol for Encrypted Execution Environments (E3s). Independent
parties use it to get one shared, verifiable result from their private inputs. This page is for
readers who are new to the protocol. It explains the problem, what the protocol does, the guarantees
and what backs each one, and the roles. At the end, you know which part of the docs to read next.
The problem
Many applications need one result from inputs that belong to different parties. A vote tally, a
sealed-bid auction, and a shared statistic are examples. Usually, one party collects the plaintext
inputs and computes the result. That party then has custody of sensitive data, and every other party
must trust it.
Trusted hardware is another approach, but it moves the trust to a hardware vendor. The Interfold
needs neither a custodian nor trusted hardware. No single party holds the plaintext inputs or the
full decryption key. Integrity and privacy come from cryptography and from economic incentives.
What the protocol does
The Interfold contract (
Interfold.sol
) coordinates each E3 on-chain. A committee of ciphernodes
does the cryptographic work off-chain and publishes proofs on-chain. One E3 has these steps:
A requester pays a fee and requests an E3 for a registered E3 program.
Sortition selects a committee of ciphernodes. A Chainlink VRF v2.5 random value seeds the
selection.
The committee runs distributed key generation (DKG) and publishes one public key. No member holds
the matching secret key.
Data providers encrypt their inputs under that public key and submit them to the E3 program.
A compute provider runs the program computation over the ciphertexts. It publishes an encrypted
output and a proof that the computation is correct.
At least T + 1 committee members publish decry프로젝트로 돌아가기