Privacy-Preserving Attribution Standards for Web Advertising

Abstract 

The W3C’s Private Advertising Technology Working Group (PATWG) is working on standardizing a Privacy-Preserving Attribution browser API to support web advertising. The PATWG is currently drafting the Level 1 version of the spec which supports advertiser measurement queries. A Level 2 version of the spec is still being designed to support more advanced queries, including ad-tech optimization queries. 

In this talk, we will present a proposal for how the Level 2 version could work, called the Hybrid proposal, which incorporates design elements from Meta and Mozilla’s Interoperable Private Attribution (IPA) proposal, Apple’s Private Ads Measurement (PAM) proposal and Google’s Attribution Reporting API (ARA). We will give some examples of realistic queries that we would like to support in both Level 1 and Level 2 including an approach to privately training a logistic regression model. 

Multi-Party Computation (MPC) is considered necessary by multiple browsers to secure the server-side processing of queries. We will present an overview of the 3-party honest-majority, malicious MPC protocol we have implemented for the Hybrid proposal and share performance results for scaling to queries with up to 2 billion reports. It is still unclear in the PATWG which MPC threat models (between 2-party malicious and 3-party honest-majority malicious) will be acceptable to major browsers. We will conclude by highlighting some queries for which we would like to gain a better understanding of the best that can be done to implement them in the 2-party and 3-party settings.