Introduction
Transport Layer Security (TLS) is the backbone of secure web communication, ensuring privacy between a client and a server. However, TLS lacks a native mechanism for a client to prove the authenticity of received data to a third party—such as a smart contract or credit agency—without sharing login credentials or relying on complete trust. Previous state-of-the-art solutions like DECO addressed this using "maliciously secure" Two-Party Computation (2PC). While effective, these methods are computationally heavy, requiring massive data transfers and high latency, which limits their practical application.
This paper introduces a novel "garble-then-prove" technique that achieves a 14x reduction in communication and an order-of-magnitude speedup, making web data authentication significantly more efficient.
The Garble-Then-Prove Paradigm
The core innovation of this work is shifting from real-time prevention of cheating to post-execution verification. In traditional malicious 2PC, the protocol must ensure the Prover cannot cheat at every step. In the Garble-Then-Prove framework, the Prover sends a standard garbled circuit to the Verifier. Once the TLS session concludes and the session secrets are no longer useful for attacking the server, the Prover reveals these secrets. A Zero-Knowledge (ZK) proof is then used to demonstrate that the initial garbled circuit was constructed correctly. This "hindsight" verification is substantially cheaper than maintaining malicious security during the computation.
Protocol Optimizations
To further enhance performance, the authors implemented several TLS-specific optimizations:
- Circuit Pruning: By identifying public components of the HMAC-SHA256 process, the team reduced the circuit size for key derivation by 2.7x.
- Efficient Authentication: The computation of authentication tags (GMAC) was moved out of the main circuit using Oblivious Linear Evaluation with errors (OLEe), a primitive two orders of magnitude faster than standard Boolean circuits.
- Commitment Conversion: A new "gadget" converts internal MPC-style tags into zkSNARK-friendly commitments (like Pedersen or KZG), allowing the authenticated data to be used in further privacy-preserving proofs.
Evaluation and Real-World Performance
The protocol was evaluated using real-world web infrastructure, specifically the Coinbase and Twitter APIs, to simulate authenticating financial and social data. To test global viability, the authors benchmarked the system with Provers located in 18 cities worldwide connecting to a Verifier in California.
Key Results
- Communication Efficiency: The protocol requires only 33.3 MB of communication for a 2KB payload, a 14x improvement over DECO's 475.7 MB.
- Latency: In wide-area network (WAN) tests with 50ms latency, the protocol completes in approximately 5 seconds, compared to over 50 seconds for previous methods. As shown in the benchmarks, a user in Tokyo can prove their Coinbase balance in roughly 7 seconds total.
- Resource Usage: Peak memory usage is reduced from roughly 3 GB in DECO-style 2PC to just 150 MB, making it feasible for consumer-grade hardware.
Conclusion
By leveraging deferred verification and cryptographic optimizations, the Garble-Then-Prove framework provides a highly efficient path for authenticating web data. While currently focused on TLS 1.2, the authors have outlined a roadmap for TLS 1.3 support. This advancement significantly lowers the barrier for integrating real-world web data into decentralized applications and other privacy-preserving systems.
Summary updated