Computing on Stuff That You Don’t Even Know: An Introduction to Secure Multiparty Computation (MPC) (mrmcd26)

Computing something that depends on your and other people’s data, without ever needing to disclose your data and without you learning any other data, sounds paradoxical. Yet, modern cryptography has tools to achieve exactly that and they begin to appear in practice. I’ll shed a light on how this actually works on the inside (no heavy math required, I promise) and ramble a bit about where it could help and where it should not. Let’s talk about secure multiparty computation (MPC), a branch of cryptography that started in the ‘80s and enables us to do amazing things: If I want to compute something not only depending on data I have, but also on someone else’s data, MPC enables that even without the other person needing to disclose their data to me, and without me disclosing my data to them. In short, MPC allows multiple people to compute on their joint data while everybody keeps full control over their data and learns nothing that they shouldn’t. This has already been used to analyze the gender pay gap and workplace discrimination, run federated analysis of patient data, or do a sugar beet auction. The first part of this talk, instead of just telling you that there is some cryptography that magically allows us to compute on private data, is about showing how this is actually done. To understand how and why MPC is possible, we will have a look at one modern cryptographic MPC protocol. There will not be a lot of fancy math or cryptography; we will mostly just need additions, multiplications, and rearrange a couple equations. As the second part, let’s see where MPC could become really helpful in the real world, and also, where it really should not be used. For instance, let’s talk about Apple’s once proposed "secure" CSAM scanning approach, how this could actually be "secure" MPC, and how Apple, behind this cryptographically guaranteed security, hides much more fundamental non-cryptographic flaws. Finally, let’s have a short look at how after 40 years of MPC, today’s tools can help you to experiment around with MPC, and how to proceed from there. https://creativecommons.org/licenses/by-sa/4.0/ about this event: https://talks.mrmcd.net/2026/talk/J3GQHJ/