SysteMPC Workshop 2026

The 2nd Workshop on Systems for Secure Multi-Party Computation (SysteMPC) will be held on September 29, 2026 in Prague, Czechia. It will be co-located with the Symposium on Operating Systems Principles (SOSP) 2026.

The SysteMPC workshop brings together systems researchers and cryptographers to discuss advances in the field of secure computation and challenges of deploying cryptographic solutions in the wild.

Keynote Talk

Speaker: Yuval Ishai, Technion and AWS
Title: Computing on Encrypted Data via Secret Dual Codes

Abstract: We revisit the question of computing on encrypted data, in the following secret-key setting. A client uploads an encryption of a large input X to an untrusted server and then wishes to make an unbounded number of queries q(X) while hiding q and X from the server, using only its secret key. How efficiently can this be done and under what assumptions?

We present efficient solutions for useful special cases, including matrix-vector multiplication and private information retrieval (PIR). These solutions rely on either the standard Learning Parity with Noise assumption, in a parameter regime not known to imply public-key encryption, or new assumptions related to the hardness of learning a secret linear subspace from noisy samples. The latter assumptions yield efficiency features that no prior approach meets, including a vanishing computational overhead on the server side.

Our core idea, inspired by prior works on PIR with preprocessing, is to encode the input X and the queries q using a pair of secret dual codes, while avoiding linear algebra attacks by adding noise.

Based on joint works with Fabrice Benhamouda, Caicai Chen, Shai Halevi, Hugo Krawczyk, Tamer Mour, Tal Rabin, and Alon Rosen

Bio: Yuval Ishai is a professor of Computer Science at the Technion, Israel, currently spending a sabbatical at Amazon Web Services, New York. He is broadly interested in cryptography, focusing especially on secure multiparty computation and efficient proof systems, and is also interested in interactions between cryptography and computational complexity theory. His work has been recognized with best paper awards from the FOCS 2004, Crypto 2007, Crypto 2016, and CCS 2025 conferences, a SIAM Outstanding Paper prize, and a TCC Test of Time Award. He is a 2018 fellow of the IACR (International Association for Cryptologic Research), served as a program chair for the TCC 2011, Eurocrypt 2019, and Eurocrypt 2020 conferences, and will chair the ITCS 2027 conference.

Invited Talk

Speaker: Nora Khayata, TU Darmstadt
Title: Optimizing Compilers for Secure Multi-Party Computation

Abstract: Developing MPC compilers contain many opportunities and pitfalls for improving concrete efficiency of MPC protocol implementations. Especially when considering hybrid (or mixed-protocol) compilation, concrete efficiency is influenced by high-level program abstractions, in-memory layouts, MPC protocol assignments, and implementation specifics of the target MPC backend. This talk gives an overview on what MPC compiler development encompasses as a whole and expands on two aspects: 1) Optimizing in-memory layouts of MPC protocol implementations and 2) generating hybrid assignments for a given MPC program.

Bio: Nora Khayata is a 4th year PhD student at the Cryptography and Privacy Engineering Group (ENCRYPTO) at TU Darmstadt, Germany, under the supervision of Prof. Thomas Schneider. Her research objectives include optimizing MPC protocol implementations, using machine-learning (ML) to improve MPC compilers, and developing cryptographic protocols for attesting ML properties.

Accepted Talks

Talk #1

Speaker: Riivo Talviste, Cybernetica
Title: When the customer knows what they want from MPC: developing an MPC-as-a-Service model for the European Statistical System

Abstract: The modern production of official statistics requires integrating sensitive, multi-source microdata under stringent legal frameworks. While PETs like MPC offer the mathematical guarantees required to protect underlying data, operationalising these technologies at scale presents a significant systems challenge. Such systems must adapt to existing, deeply entrenched institutional workflows and rigorous real-world production requirements. Traditional architectures create friction in this domain because they rely on data centralisation or a single point of trust. On the other hand, typical one-off MPC pilots are too brittle and bespoke to survive in continuous production, demanding local infrastructure management from participants and accumulating prohibitive upkeep cost over time.

To bridge this gap, the JOCONDE project procured by Eurostat designed a Multi-Party Secure Private Computing-as-a-Service (MPSPCaaS) solution tailored to augment the existing production processes of the European Statistical System (ESS). JOCONDE aims to replace isolated, single-use pilots with a reusable infrastructure utilising a centrally managed deployment that services all ESS members without establishing a single point of trust. Operational management is centralised to match real-world usability needs, yet cryptographic and organisational trust boundaries ensure data remains completely protected from all participants, including the solution administrators themselves.

Crucially, JOCONDE abstracts cryptographic complexity into an on-demand, multi-tenant service layer that integrates seamlessly into day-to-day statistical workflows. The architecture employs an MPC-over-TEE hybrid model for defence-in-depth, satisfying strict institutional performance and security mandates. Finally, to eliminate potential vendor lock-in, JOCONDE solution decouples the core application layer from the underlying MPC computation engines through a highly modular interface design. This allows onboarded participants to dynamically swap MPC backends and even TEE hardware over time.

By sharing the system specifications and architectural design choices from the JOCONDE project, this presentation provides a concrete blueprint for transforming MPC from a series of isolated pilot exercises into a reliable, reusable, and continuous infrastructure for public interest computing. We also touch on the extensive regulatory and legal work conducted in the project.

Bio: Riivo Talviste is a senior security engineer at Cybernetica AS, where he has been working for 15 years on the practical aspects of MPC. He has been at the forefront of several MPC deployments that work on real data. Riivo received his PhD from the University of Tartu, with his research centred on the practical lessons learned from applying MPC in real-world scenarios.

At Cybernetica, Riivo leads the development of Sharemind MPC, one of the earliest secure multi-party computation platforms. In recent years, the development focus has been on long-term maintainability and embracing cloud-native technologies.

Talk #2

Speaker: Shailesh Mishra, EPFL
Title: Tutela: Privacy-Preserving Biometric Deduplication at Scale

Abstract: Biometrics offer a natural means of enforcing one identity per person, owing to their uniqueness, ease of use, and permanence. However, biometric processing inherently raises privacy concerns and jurisdictional constraints on sharing data across regions. Hence, we present Tutela, a privacy-preserving biometric deduplication system that simultaneously (i) keeps biometric templates inside their region of origin, (ii) leaks nothing about a region’s database beyond a single duplicate / no-duplicate bit per cross-region check, and (iii) scales to populations of hundreds of millions—three requirements that no existing private ANN system meets together. Tutela partitions infrastructure along legal-geographic regions: each region operates its own MPC nodes and PIR servers, and templates never leave their region. At its core is HiT-ANN, a new approximate nearest-neighbor structure tailored for cryptographic execution: HiT-ANN is inspired by Thorup–Zwick compact routing schemes to represent the database as a hierarchy of small, bounded-size hop tables, and we believe it to be of independent interest. On top of HiT-ANN, Tutela uses a composite MPC + PIR sub-protocol: secure multi-party computation (MPC) finds the closest entry within each hop table, while multi-server private information retrieval (PIR) obliviously fetches the next layer’s hop table from the target region. This MPC-PIR division of computation avoids the cost of running all of HiT-ANN’s search under MPC, which reduces the bandwidth and latency significantly. We implement and evaluate a prototype of Tutela. Our techniques — HiT-ANN’s bounded-size hop tables and the composite MPC + PIR search protocol — ensure that Tutela achieves an order of magnitude better performance than a linear MPC baseline for privacy-preserving cross-region deduplication.

Bio: Shailesh Mishra is a fourth-year Ph.D. candidate at the School of Computer and Communication Sciences, EPFL, Switzerland, being advised by Prof. Bryan Ford in the Decentralized and Distributed Systems (DEDIS) Laboratory. His research interests are in all aspects of privacy-preserving digital identities, particularly private account recovery using secret sharing, privacy-preserving deduplication, and accountability with privacy. Previously, he received a Bachelor’s + Master’s in Electrical Engineering from IIT Kharagpur, India, in 2022.

Talk #3

Speakers: Pierfrancesco Ingo and Vaastav Anand
Title: Demystifying the multiverse of MPC protocols

Abstract: Secure multiparty computation (MPC) enables mutually distrusting parties to compute over private inputs with strong privacy and correctness guarantees. However, despite many proposed applications, real-world adoption remains limited by high costs (especially in the malicious setting) and lack of guidance on choosing suitable protocols for concrete workloads and deployment settings. Consequently, developers face a sprawling design space spanning threat models, corruption thresholds, computation domains, preprocessing strategies, and protocol families—each with very different performance characteristics.

In this talk, we demystify this “multiverse” of MPC protocols and provide a practical map of the design space. The goal of the talk is to give both newcomers and practitioners an intuition for why MPC protocols perform differently and a framework for choosing among them without having to become an expert in every corner of the MPC literature. To do so, we first explain the key dimensions that distinguish general-purpose MPC protocols and how they translate into performance trade-offs. We then present an extensive empirical comparison using MP-SPDZ, benchmarking representative protocol families across four common computational primitives. Our results show that there is no universally best MPC protocol. Instead, the right choice depends strongly on the structure of the workload and the deployment environment. Finally, we conclude with practical guidelines for selecting MPC protocols and discuss what these results suggest for future MPC frameworks, benchmarking, and automatic protocol selection.

Author List: Roberta De Viti, Vaastav Anand, Pierfrancesco Ingo, Deepak Garg. For more info, please refer to our arXiv paper.

Bios: Pierfrancesco Ingo is a researcher specializing in distributed systems security, with a particular interest in differential privacy, differential anonymity, and secure multi-party computation. His work focuses on formalizing probabilistic models for mixnets and developing efficient systems based on secure multi-party computation. He has experience in privacy-enhancing technologies, cryptographic protocols, distributed systems, and the design and analysis of privacy-preserving systems.

Vaastav Anand is a researcher specializing in cloud systems reliability and optimization. His work focuses on building human-in-the-loop specification-driven automation tools for cloud system design, optimization, maintenance, and improvement.

Talk #4

Speaker: Ioannis Demertzis, UC Santa Cruz
Title: Obliviousness Is the Common Denominator: Scalable Secure Oblivious Primitives for TEE and MPC

Abstract: Trusted execution environments and secure multi-party computation are usually presented as competing approaches to computing on private data: hardware trust versus cryptographic trust, near-native speed versus provable guarantees under a non-collusion assumption. This talk argues that this framing hides what the two stacks share. A decade of leakage-abuse attacks on encrypted databases and side-channel attacks on TEEs tells the same story: encryption alone is not enough, because memory access patterns, page faults, cache behavior, and result sizes reveal the data. TEEs therefore need their algorithms to be made oblivious, so that observable behavior is independent of the input. MPC is oblivious by construction and pays worst-case costs for it. In both worlds the problem is the same, designing data-independent computation that scales, and so is the toolkit: oblivious sort, shuffle, compaction, key-value stores and indexes, and the database operators and query plans built on top of them. What differs is only the cost model.

Drawing on our work on deployed oblivious systems, from Snoopy (SOSP’21), the oblivious object store behind Signal’s private contact discovery at 100M+ user scale, to distributed oblivious sort and shuffle (S&P’24), oblivious parallel database operators (OBLIVIATOR, USENIX Security’25), concurrent oblivious stores scaling to hundreds of terabytes (SONIC, USENIX Security’26), and oblivious multi-way joins (DISCO, SOSP’26), I will focus on a set of design principles that transfer across the TEE/MPC boundary. These primitives already underpin real-world deployments, from Signal and Apple’s Private Cloud Compute to Google’s Privacy Sandbox and AWS Clean Rooms, and the same building blocks are now poised to power private databases, analytics, and AI serving across both trust models.

Bio: Ioannis Demertzis is an Assistant Professor in the Computer Science and Engineering Dept. at the University of California, Santa Cruz. His research focuses on applied cryptography, security & privacy, and secure databases/systems. His work has been published at top security, systems, and database conferences, including USENIX Security, CRYPTO, S&P, SIGMOD, SOSP, and PVLDB. His research is supported by an NSF CAREER Award, an Amazon Research Award, and research grants from the Stellar Development Foundation, the Sui Foundation, and Nillion. He is also the recipient of the ACM SIGSAC Doctoral Dissertation Award Runner-up, the Distinguished Dissertation Award of ECE (University of Maryland), and the Symantec Research Labs Graduate Fellowship. Before joining UCSC, he was a Postdoctoral Researcher at the EECS Dept. of UC Berkeley. He received his Ph.D. from the ECE Dept. of the University of Maryland, College Park, and his ECE Diploma and M.Sc. from the Technical University of Crete.

Accepted Posters

  1. Extensible Query Optimization for the Privacy-Performance Trade-off in Secure Collaborative Analytics
    Bio: Dr.-Ing. Shaza Zeitouni – PhD in secure systems from the Technical University of Darmstadt with a focus on novel hardware‑rooted trust mechanisms and their integration into embedded platforms. Since 2024, he has been a post‑doctoral researcher at the Systems Group, leading research on privacy‑preserving data analytics. Her current work tackles the scalability bottleneck of Secure Multi‑Party Computation (MPC) for OLAP workloads.
  2. Neonik: MPC evaluations made easy
    Bio: Vincent Ehrmanntraut is a third-year PhD candidate at RWTH Aachen University. His research focuses on MPC protocols for solving optimization problems arising in the context of circular economies, particularly graph problems such as maximum flow.
  3. Beyond Worst-Case Padding: Materializing Multi-Way Joins in MPC
    Speaker: Apostolos Mavrogiannakis, UCSC

Call for papers

Topics of interest

  • Cryptography and systems co-design
  • Programming abstractions for MPC
  • Performance optimization for secure computations
  • Secure query optimization
  • Domain-specific systems for MPC, e.g., secure machine learning, graph analytics, time series analytics
  • Integration of MPC with other PETs
  • Modularity of cryptographic software
  • Hardware acceleration
  • Operating systems optimizations for MPC
  • Experimentation, benchmarks, and evaluation studies
  • Deployment and orchestration of MPC systems
  • Integration of MPC tools with the existing ecosystem

Important Dates

  • Submission deadline: July 10, 2026
  • Notification deadline: July 24, 2026
  • Workshop date: September 29, 2026

Talk and Poster Submission

We invite submissions for full talks (20 min), lightning talks (5 min), and posters. Both recently published works and works-in-progress are welcome.

Submission form: https://forms.gle/R52xGLCL5gTRAoZ26

Organizers

This half-day workshop is organized by Vasia Kalavri, John Liagouris, and Mayank Varia from the BU systems research group and the BU security group.