Bishakh Chandra Ghosh (Indian Institute of Technology Kharagpur), Sikhar Patranabis (IBM Research - India), Dhinakaran Vinayagamurthy (IBM Research - India), Venkatraman Ramakrishna (IBM Research - India), Krishnasuri Narayanam (IBM Research - India), Sandip Chakraborty (Indian Institute of Technology Kharagpur)
We initiate the study of Private Certifier Intersection (PCI), which allows mutually distrusting parties to establish a trust basis for cross-validation of claims if they have one or more trust authorities (certifiers) in common. This is one of the essential requirements for verifiable presentations in Web 3.0, since it provides additional privacy without compromising on decentralization. A PCI protocol allows two or more parties holding certificates to identify a common set of certifiers while additionally validating the certificates issued by such certifiers, without leaking any information about the certifiers not in the output intersection. In this paper, we formally define the notion of multi-party PCI in the Simplified-UC framework for two different settings depending on whether certificates are required for any of the claims (called PCI-Any) or all of the claims (called PCI-All). We then design and implement two provably secure and practically efficient PCI protocols supporting validation of digital signature-based certificates: a PCI-Any protocol for ECDSA-based certificates and a PCI-All protocol for BLS-based certificates. The technical centerpiece of our proposals is the first secretsharing-based MPC framework supporting efficient computation of elliptic curve-based arithmetic operations, including elliptic curve pairings, in a black-box way. We implement this framework by building on top of the well-known MP-SPDZ library using OpenSSL and RELIC for elliptic curve operations, and use this implementation to benchmark our proposed PCI protocols in the LAN and WAN settings. In an intercontinental WAN setup with parties located in different continents, our protocols execute in less than a minute on input sets of size 40, which demonstrates the practicality of our proposed solutions.