Large-Scale Private Computation for Real-World Applications via Trusted Hardware and Obliviousness

Privacy-preserving computation that exposes memory access patterns can inadvertently reveal significant information about plaintext inputs through leakage-abuse attacks. While industrial approaches utilizing hardware enclaves for confidential computing are vulnerable to side-channel attacks, they offer an affordable and cost-effective solution for any privacy-preserving computation. Oblivious primitives, as robust cryptographic tools, can mitigate leakage-abuse and software side-channel attacks (when integrated with hardware enclaves). In this tutorial, we present the latest advancements in TEE-based oblivious primitives, highlighting multiple applications such as private contact discovery for Signal, our new proposal for anonymous communication resistant to long-term traffic analysis (SP'25), oblivious relational (USENIX'25) and graph databases (PVLDB'24), key transparency, searchable encryption, large-scale software activity monitoring, federated learning, privacy-preserving LLMs, Google's Privacy Sandbox initiative, Google's FLEDGE, Google's Titan Security Key and Google's Asylo. Our recent state-of-the-art oblivious primitives include a high-throughput oblivious key-value store (used in Signal's commercial solution for contact discovery, SOSP'21), a reduced-latency solution (PVLDB'24), the most scalable oblivious sort and shuffle primitives to date (SP'24), optimized for shared-memory and distributed architectures, and the first scalable oblivious filter, group-by, join approaches (USENIX'25).