Skip to content

Security and privacy model for ad confirmations

Jump to bottom
Francois Marier edited this page Jun 27, 2024 · 9 revisions

Application

Parties

  • Users running Brave browser
  • Advertisers
  • Brave operators

Resources

  • Internet
  • User computers, Brave browser installation with stored secrets
  • Challenge bypass microservice
  • Ads server
  • Ledger server (wallet service)
  • Eyeshade server (accounting service)

Goal

  • Given a Brave user observes an advertisement, provide a channel for "confirming" said viewing without revealing to Brave the particular user involved.
  • Upon receiving a valid confirmation, grant the user a redeemable payment token that is worth ~70% of the observation price charged to the advertiser.
  • Allow a user to redeem their payment tokens, this redemption should be unlinkable to the original confirmation events.
  • Pay out for redeemed tokens on-demand after offering a notification to the user to claim their earnings.
  • Prevent double spend attacks by ensuring each payment token can be spent at most once.

Specific user privacy goals

  • Brave/Advertiser cannot learn that a user in Segment X and a user in Segment Y are the same user (AKA, user unlinkability across segments)
  • Brave/Advertiser cannot learn that a user who has viewed or interacted with Creative Set X and a user who has viewed or interacted with Creative Set Y are the same user (AKA, user unlinkability across creative sets)
  • Brave/Advertiser cannot figure out any personally-identifiable or sensitive information about the user interacting with an ad creative without explicit consent of the user.

"Personally-identifiable or sensitive information" in this context includes but is not limited to: the user's browsing history, their email address and contact info, their real name, their Brave payments history, what ads they have seen, what segments they are in, their payments wallet address and anything considered personal info under GDPR.

Threat model

  • MITM on internet between user and ads server
  • Crash / power loss
  • Timing based loss of privacy
  • Malicious advertisers
  • Fraudsters interacting with ads server using the Brave browser or another client

Attack surfaces

  • HTTPS from the world over internet to CDN
  • HTTPS from CDN to the ads server
  • HTTPS from the ads server to the challenge bypass microservice
  • HTTPS from the ads server to the eyeshade server
  • HTTPS from the ads server to the ledger server

Out of scope:

  • Compromise of a user's device
  • Brave collusion with CDN to track users based on IP

Protocol

The ads confirmation protocol is composed of two separated rounds of the Privacy Pass cryptographic protocol, of which you can find a brief description here.

Server Setup

  1. The ads server communicates with the challenge bypass microservice to create an issuer (consisting of a particular SigningKey) for confirmations.
  2. The ads server communicates with the challenge bypass microservice to create a small number of issuers for payment. Each key corresponds to a particular payout value, for example there could be keys for 0.1, 0.3, 0.5, 1, 3, and 5 BAT.
  3. The corresponding PublicKeys for the confirmations and payments are published in the ads catalog.

Client Setup

  1. A user generates a pool of Tokens, blinds them and submits a HTTP signed request to the ads server to have them signed by the server's confirmation SigningKey.
  2. The ads server makes a request to the ledger server to retrieve the wallet public key, checks the signature over the request and finally signs the tokens. Furthermore a BatchDLEQProof is created demonstrating that the tokens were signed by a SigningKey whose PublicKey was previously published in the catalog.
  3. The returned BatchDLEQProof is verified and the SignedTokens are unblinded and stored locally in their confirmation token pool.

The HTTP signature over this request restricts confirmation token issuance to valid rewards wallets. A SigningKey rotation can be performed to force clients to create a new pool of tokens.

Upon viewing an ad

  1. A user pops an UnblindedToken from the pool, uses it to form a VerificationKey and uses the key to "sign" confirmation metadata.
  2. The user forms a token redemption request consisting of the VerificationSignature, the request metadata, the TokenPreimage associated with the UnblindedToken, and the PublicKey corresponding to the key it was signed by.
  3. The user generates a new Token, and blinds it. Together with the token redemption request this forms a confirmation request.
  4. The confirmation request is submitted to the ads server, the client receives a unique identifier by which it can later retrieve a SignedToken.
  5. Using the included PublicKey the ads server looks up the issuer
  6. The ads server forwards the token redemption request to corresponding issuer endpoint of the challenge bypass microservice, which marks the TokenPreimage as spent.
  7. The ads server rounds the value of the confirmed event to the nearest value that has a corresponding payout key, then uses this key to sign the BlindedToken. A BatchDLEQProof is also created showing the payment token is signed by a PublicKey in the catalog.
  8. The returned BatchDLEQProof is verified and the returned SignedToken is unblinded and stored locally with the other payment tokens the user has received.

These confirmation token redemption events should not be linkable to the user to whom they were issued provided:

  1. There is a sufficient anonymity set formed by users who are creating confirmation tokens.
  2. The time between token issuance and confirmation is not predictable.
  3. The confirmation metadata does not include information unique to the user (that is not otherwise unique to this event.)
  4. The user is not made re-identifiable through other side channels such as IP address information. We ensure this by configuring our CDN to neither log nor forward client address information on the anonymous confirmation endpoint.
  5. The signing keys used to sign both confirmation and payment tokens are widely used, we do not use unique keys for small user cohorts. To this effect we publish all signing keys publicly and instruct the browser to verify DLEQ proofs.

Weekly

  1. A user takes all stored payment tokens that they have accumulated and forms a bulk token redemption request, submitting their wallet identifier as the metadata that is "signed".
  2. Using the included PublicKey for each token redemption request the ads server looks up the issuers for each - and thus the corresponding value.
  3. The ads server forwards the token redemption requests to the corresponding issuer endpoints of the challenge bypass microservice, which marks the TokenPreimage of each as spent and ensures that a double spend has not occured.
  4. The ads server makes a request to the eyeshade server, submitting a request for payment for the value of each token.

During the first week of the month

  1. The browser client will request payment against all of the collected confirmations
  2. A Brave operator will pull the balances against the confirmations to be paid - only when the confirmations are received, earnings are earmarked for the client
  3. Ad earnings are then made available for users to claim
Clone this wiki locally