PEPR '26 - Private AI: Building Trust Through Verifiable Computation @UsenixOrg

Uploaded June 2026 | Updated September 2026, 3 weeks ago

Private AI: Building Trust Through Verifiable Computation

Mingshen Sun and Mateus Guzzo, TikTok

AI has transformed how people learn, work and live - automating complex tasks and extracting insight from massive datasets. But most powerful AI today (especially large language models) runs on server-class hardware, which typically means user prompts and context must be visible to the service provider to be processed. While acceptable for some cases, it is still challenging with highly sensitive data where users expect similar protections as end-to-end encryption. Private Verifiable Compute (PVC) is a technical solution that can enable users to initiate a request to a private and verifiable environment for context-aware AI processing with sensitive data, where no one, including service providers, can access them. With PVC in the cloud environment, it unleashes full potentials of AI hardware in the data center for complex AI tasks, such as large language models (LLMs), generative AI and beyond, while guaranteeing user privacy and verifiable transparency.

View the full PEPR '26 program at usenix.org/conference/pepr26/program

PEPR 26 - Private AI: Building Trust Through Verifiable Computation

SREcon26 Americas - Building SRE Culture (without SREs, Technically)

$NSDI 26 - Detecting and Diagnosing Errors in Serving Archived Web Pages Detecting and Diagnosing Errors in Serving Archived Web Pages Jingyuan Zhu, University of Michigan; Huanchen Sun and Harsha V. Madhyastha, University of Southern California Community Award Winner! Web archives crawl and save copies of pages from the web, enabling users to interact with web pages in the form they existed in the past. Prior to serving any archived page, an archive rewrites the page’s source so that users’ browsers fetch the page’s resources from the archive, not from the servers which originally hosted them. But, on many modern pages, an archive’s edits to crawled scripts result in a loss of fidelity, i.e., an archived copy fails to accurately mimic the original page even when the archive had crawled all resources on the page. To help the developers of archival systems identify and fix the bugs which result in incorrect rewrites of crawled pages, we present FidEx. First, FidEx enables accurate identification of the pages on which an archive violates fidelity. It does so by tracking and comparing the execution of scripts between when a page is crawled and when its copy is loaded. In comparison to existing methods which compare the two loads using either screenshots or the errors reported by the browser, FidEx reduces the false positive rate from around 70% to less than 10%. Second, on every page on which it identifies a loss of fidelity, FidEx pinpoints which subset of the archive’s edits to the page are erroneous. Leveraging this input to fix bugs in the most widely used archival system, we reduced the fraction of archived pages which violate fidelity from 15% to 9%. View the full NSDI 26 program at https://www.usenix.org/conference/nsdi26/technical-sessions$

NSDI 26 - SYMI: Efficient Mixture-of-Experts Training via Model and Optimizer State Decoupling

PEPR 26 - Vision: Human-as-the-Unit Privacy Management with AI Agents

PEPR 26 - Surfacing Hidden Privacy Risks in Code: Lessons from LLM and Retrieval Assisted Detection

NSDI 26 - The GOODPUT System: A Machine Learning-Driven Optimization Framework for Dynamic...

NSDI 26 - PrvTel: Lightweight Models for Private and Accurate Telemetry Data Retention

NSDI 26 - Count-Based Abstractions for Performance Verification of Contention Points

NSDI 26 - A Systematic Threat Analysis and Practical Attacks on Automated Frequency Coordination

NSDI 26 - ForestColl: Throughput-Optimal Collective Communications on Heterogeneous Network Fabrics

NSDI 26 - CacheCatalyst: Enhancing Web Caching for the Latency-Constrained Internet