NDSS 2026 Showcase: Hardcore Cryptography (Day 4)

We have reached the final day of my NDSS 2026 Showcase. NDSS is one of my most favorite parts of my job at Internet Society and I want to thank, up front, my fantastic team, including Robin Wilton, Raquel Kroich, Sally Harvey, Ivana Strineka Trbovic (I'm forgetting people, I know it)!!!

So far this week, we have explored physical-layer "Mad Science" hacks, the societal impact of "Digital Rights", and the fragility of "Internet Infrastructure." For Day 4, we end with a sense of awe at the mathematical ingenuity driving our field forward. This theme, Hardcore Cryptography, covers the breakthroughs that will define the future of privacy and secure computation. These papers take theoretical concepts—like zero-knowledge proofs and homomorphic encryption—and make them faster, scalable, and usable in the real world:

1. VeriLoRA: Fine-Tuning LLMs with Verifiable Security

Liao et al. present a framework that solves a massive trust problem in AI: How do you know a company actually trained a model the way they said they did? VeriLoRA uses Zero-Knowledge proofs to allow companies to generate a "digital receipt" verifying that a Large Language Model (LLM) was fine-tuned correctly. Crucially, this proof confirms the integrity of the update without ever revealing the private model weights or the sensitive training data. Read the paper here

2. Cirrus: Performant and Accountable Distributed SNARKS

Wang et al. tackle the bottleneck of generating privacy proofs, which usually requires massive computing power. They introduce Cirrus, a system that distributes the workload of generating zk-SNARKS across many different computers. To prevent cheating, the system includes a cryptographic "lie detector" that enforces accountability, instantly catching and punishing any worker node that attempts to submit a fraudulent computation. Read the paper here

3. Select-Then-Compute: FHE for Distributed Datasets

Koirala et al. propose a major step forward for collaborative research in sensitive fields like healthcare. Using Fully Homomorphic Encryption (FHE), they developed a framework that allows researchers to filter and analyze encrypted data across distributed locations (like different hospitals). This "Select-Then-Compute" approach enables complex analytics—such as studying patient trends—without the researchers ever seeing the raw files or unlocking the data. Read the paper here

4. MinBucket MPSI: Breaking the Max-Size Bottleneck in Private Set Intersection

Tu et al. introduce a protocol that dramatically speeds up the process of comparing secure datasets. Traditionally, if a small company wanted to find common customers with a giant company, the process was slow because the larger dataset bottlenecked the computation. MinBucket MPSI uses hashing buckets to break lists into tiny chunks, decoupling performance from dataset size and making private set intersection lightning-fast even for massive lists. Read the paper here

This wraps up our highlight of 17 papers from NDSS 2026. However, this is just a fraction of the work being presented; there are over 200 more papers and 8 workshops to explore. The quality of research at NDSS is a testament to the community's dedication to a secure, open, and trustworthy Internet. Please visit the NDSS website, register to attend, and consider donating to the Internet Society to keep this research free for all.