Open-sourcing Nameless Credential Service

• Meta has open-sourced Nameless Credential Service (ACS), a extremely out there multitenant service that permits purchasers to authenticate in a de-identified method.
• ACS enhances privateness and safety whereas additionally being compute-conscious.
• By open-sourcing and fostering a neighborhood for ACS, we consider we are able to speed up the tempo of innovation in de-identified authentication.

Information minimization — gathering the minimal quantity of knowledge required to help our providers — is considered one of our core ideas at Meta as we develop new privacy-enhancing applied sciences to guard person knowledge on our household of merchandise. The purpose is to ship worthwhile person experiences whereas gathering and utilizing much less knowledge. 

Our method to logging is one essential instance of this follow. Logging helps our engineers and builders consider efficiency and reliability, enhance product options, and generate stories.

Consumer identities aren’t needed in most logging use instances and needs to be excluded from logging knowledge. Eradicating authentication is one option to take away identifiers. However doing so makes the system weak to varied assaults, together with knowledge injection. 

At Meta, we’ve constructed a greater method for purchasers to authenticate in a de-identified method: Nameless Credential Service (ACS). At a excessive degree, ACS helps de-identified authentication by splitting authentication into two phases, token issuance and token redemption. Within the token issuance part, purchasers contact the server by an authenticated channel to ship a token. The server indicators the token and sends it again. Then, within the de-identified authentication (or token redemption) part, purchasers use a de-identified channel to submit knowledge and authenticate it using a mutated type of the token slightly than a person ID.

ACS has performed an essential function in how we do de-identified authentication at scale. Now we’ve open-sourced it so the bigger neighborhood can each profit from ACS and assist speed up innovation in de-identified authentication.

Right here’s how we developed ACS, and how one can get began utilizing it.

An outline of the nameless credential protocol

The nameless credential protocol is constructed on prime of verifiable oblivious pseudorandom capabilities (VOPRFs) and blind signatures. 

Taking logging for example once more, we clear up the issue of de-identified logging by splitting the workflow into two steps: First, purchasers use an authenticated connection to the server to acquire an nameless credential upfront. Then, every time the purchasers must add logs, they ship the nameless credential together with the logs in an unauthenticated connection to the server. The nameless credential serves as proof that the consumer is genuine.

Right here’s how the method performs out:

Step 1 (token issuance):

1. The consumer generates a token.
2. The consumer blinds the token.
3. The consumer sends the blinded_token to the server, together with authentication knowledge.
4. The server indicators the blinded_token after which sends the signed_blinded_token again to the consumer.
5. The consumer unblinds the obtained token, leading to a signed_unblinded_token.

Step 2 (token redemption):

1. The consumer sends the unique token, signed_unblinded_token, together with the enterprise knowledge it wants for the use case (e.g., logging occasions) to the server.
2. The server validates the request with tokens. If the consumer is genuine and licensed to entry, the server will course of the enterprise knowledge.

This protocol is efficient as a result of:

• The enterprise knowledge and authentication knowledge are separated.
  • The enterprise knowledge is shipped with unblinded tokens, and authentication knowledge is shipped with a blinded token. It’s noteworthy that the token issuance step and token redemption step don’t occur on the similar time — the consumer can retailer tokens for a number of hours and even a number of days. If the consumer needs to log knowledge however is out of tokens, they’ll fetch a token and redeem it instantly. However these two steps are put into separate requests to assist stop an id from being inferred from the information.
• The token, along with signed_unblinded_token, serves because the legitimation of the consumer. The token issuance server makes use of a secret key to signal tokens, and that secret key can’t be inferred from client-side observations (see: decisional Diffie–Hellman assumption). 

Challenges of the nameless credential protocol

To make the protocol work in real-life, large-scale methods, there are extra challenges to be solved.

Token redemption counting

Ideally, one credential will be redeemed solely as soon as. However in follow, it’s acceptable to permit a credential to be redeemed a number of instances (as outlined by the use case) to cut back server load. We utilized a real-time, dependable, and secured counting service to restrict the variety of token redemption instances.

Key rotation

The nameless credential protocol requires a key pair. The server makes use of a secret key to signal the token (step 1.4) and validate the redemption request (step 2.2). The consumer wants a corresponding public key to unblind the token (step 1.5).

Given this, key administration — particularly, rotating keys often and discarding stories from previous keys — performs a vital function in making certain that we are able to mitigate the impression of purchasers if they’re compromised after they’re issued a credential. These key rotations should be deployed throughout the fleet in a constant  and environment friendly method. The important thing administration service interacts with the configuration administration system to mutate key supplies for ACS tenants in keeping with the cipher suites and key rotation schedules specified of their configuration recordsdata. 

There are additionally challenges round distributing new verification keys to purchasers that want to confirm credentials.

Key transparency and attribute-based VOPRFs

The design of our attribute-based VOPRFs is motivated by our want for an environment friendly and clear technique round key rotation.

Frequent key rotations present a safety measure for ACS. Nonetheless, a malicious server can determine customers by signing every one with a user-specific key that may be tied again to them throughout credential redemption. 

Key transparency makes it doable for customers to learn about all of the out there public keys, stopping the server from assigning user-specific key pairs. Furthermore, at Meta we have to handle many keys for every ACS use case, and sustaining naively generated keys will not be scalable.

We solved this downside by introducing key derivation capabilities (KDFs). At a excessive degree, given any attributes (e.g., a bunch of strings), new secret keys will be derived from public keys, which may additional be derived from a single public key. By setting the attributes to discuss with the time epoch for which the keys are legitimate, purchasers will be verified simply with out the necessity to fetch new public keys.

Because of this, we are able to prolong the transparency of the first public key — which will be shipped with consumer code or posted to a trusted location — to those derived public keys with none extra effort.

Deploying nameless credential protocol at scale

With these concerns in thoughts, a typical ACS deployment appears to be like extra like:

Setup (step 0):

1. The consumer obtains the server’s major public key and different public parameters.
2. The server generates a key pair utilizing given attributes (use case title, time epoch, identified to purchasers) after which sends the general public key to the consumer.
3. The consumer validates the general public key with the first public key and attributes.

Step 1 (token issuance):

1. The consumer generates a token.
2. The consumer blinds the token.
3. The consumer sends the blinded_token to the server, together with authentication knowledge.
4. The server checks the token issuance price for the precise person. It then indicators the blinded_token and sends the signed_blinded_token again to the consumer.
5. The consumer unblinds the obtained token, leading to a signed_unblinded_token. 

Step 2 (token redemption):

1. The consumer sends the unique token, signed_unblinded_token, together with the enterprise knowledge it wants for the use case (e.g., logging occasions) to the server.
2. The server validates the request and checks the redemption instances for the precise token. If the consumer is genuine and licensed to entry, the server will and course of the enterprise knowledge. 

Step 0.3 performs an essential function in sustaining key transparency. If a malicious server is assigning public keys that correlate to person authentication knowledge, the validation step would fail and the consumer may refuse to make use of the general public key obtained.

Learn the paper “DIT: De-identified authenticated telemetry at scale” for extra mathematical particulars for the protocol.

The ACS library

The ACS repo offers a portal and extensible C library (within the /lib/ folder), whose important parts embody:

• The VOPRF protocol: This contains client-side token blinding, unblinding, and producing a shared secret for token redemption. For servers, the protocol contains signing the blinded token and producing a server-side shared secret for token redemption. There are two variations of the blinding methodology offered within the library.
• An attribute-based key derivation operate: It is a key rotation resolution. If the attributes are set to a standard identified worth (e.g., time epoch), purchasers can confirm the authenticity of the server simply. There are a number of KDFs offered within the library. We advocate Sturdy Diffie–Hellman Inversion (SDHI) or Naor-Reingold for higher key transparency.
• Discrete log proof: That is used to show the authenticity of the server. It’s used twice within the protocol — first, to confirm the general public key derived from attributes within the setup step, and second, to confirm the signed token in token issuance step
• Elliptic curves: The ACS library is modular, and customers can select most popular elliptic curves. Ed25519 and Ristretto are at the moment offered.     

The library is meant to be deployed on cell units, so we wish to reduce exterior dependencies to maintain the binary measurement small. Presently, libsodium is the one dependency for the ACS library.

Along with that, we have now carried out a SimpleAnonCredService (server + consumer) in C++ for demonstration functions. The service is constructed with Apache Thrift 0.16. (See the /demo/ folder within the repo.)

How you can use ACS in an actual system

Let’s use an instance to reveal the workflow. Suppose we’re sustaining a service that permits authenticated customers to get climate stories. A naive system will seem like this:

# consumer
get_report(authentication_data)
# server
if check_authentication(request.authentication_data):
    response.report = report_data

Step one is to separate the authentication_data from report_data, which is the principle goal of the ACS undertaking. 

# consumer - authentication
token = random_string()
blinded_token, blinding_factor = blind(token)
signed_blinded_token = request_token_from_server(authentication_data, blinded_token)
signed_unblinded_token = unblind(signed_blinded_token, blinding_factor)
# consumer - get knowledge
client_secret = client_finalize(token, signed_unblinded_token)
get_report(token, client_secret)
# token issuance server
if check_authentication(request.authentication_data):
    signed_blinded_token = consider(blinded_token)
    response.signed_blinded_token = signed_blinded_token
# token redemption server
server_secret = server_finalize(request.token)
if server_secret == request.client_secret:
    response.report = report_data

After the consumer is authenticated and requests the information it wants, the consumer generates a token, blinds the token, and sends the token to the server. After an authentication examine, the server indicators the token and sends it again to the consumer. The consumer then unblinds the signed token, after which verifies it with the general public key and proof.

Lastly, the consumer redeems the token. The server validates the key key and proceeds to enterprise logic if the validation succeeds. If the validation fails, the server rejects the request.

Once we launched key rotation and KDF, it added two extra steps at first of the method:

1. The consumer downloads the first public key from the server. This major public key’s for validation of the general public key in step 2.
2. The consumer will get a public key for offered attributes. The attributes will be any checklist of strings (e.g., use case names, dates) which are allowed by the server. KDFs enable for key transparency. After this step, the consumer will likely be assured that the server will not be assigning a public key associated to the authentication info. Later, the general public key can be utilized within the verifiable_unblind step to ensure the signed_blinded_token is signed with the non-public key equivalent to the verified public key.

# consumer - setup
primary_public_key = request_primary_public_key_from_server()
# consumer - authentication
public_key, pk_proof = get_public_key_from_server(attribute)
if !dleqproof_verify(public_key, pk_proof, primary_public_key, attribute):
    increase Exception("malicious server!")
token = random_string()
unblinded_token, blinding_factor = blind(token)
signed_blinded_token, proof = request_token_from_server(authentication_data, blinded_token)
signed_unblinded_token = verifiable_unblind(signed_blinded_token, blinding_factor, proof, public_key)

With all these steps, we’ve prevented a probably malicious server from utilizing these key rotations to segregate and determine customers. It is a good prototype system and able to use. However in a scalable system, there are extra challenges to beat, together with client-side token storage and server-side price limiting. These options will not be included in ACS’s open supply repo.

Future plans for ACS

Wanting on the future, we consider the modular ACS is extensible and has the potential to be useful to industries that make the most of nameless credential options. We’re planning to implement the customary.

A light-weight model with out libsodium dependency will likely be useful to make use of instances the place binary measurement is restricted. 

When you’d wish to contribute to the undertaking, please go to the ACS GitHub.