Loading
Life sciences · Preprint
arXiv · September 9, 2026
Early or partial results. Treat as a signal, not a conclusion.
Maverick is a novel protocol and prototype system for private and verifiable LLM inference that delegates matrix-vector multiplication with claimed information-theoretic soundness and efficient verification. Throughput benchmarks on a 4B-parameter model show speedups of up to 45x over local inference under precomputed masking, but the work is a preprint prototype lacking peer review, independent validation, and comparison against existing privacy-preserving inference approaches.
Protocol design with prototype implementation and benchmark evaluation. Evaluation on Qwen3-4B, an open-source 4-billion-parameter language model. No description of user population, real-world deployment, or adversarial settings.. Intervention: Maverick: a protocol for delegating matrix-vector multiplication in LLM inference with information-theoretically sound verification and LPN-based input privacy masking.. Compared with: Local inference (implicit baseline for throughput comparison); no explicit comparison against other privacy-preserving or verifiable inference methods..
Maverick achieves throughput gains of up to 17x with online privacy mask generation, 45x with precomputed masks, and 44x with verification-only on Qwen3-4B using one client thread and 128 CPU server threads. With four client threads, corresponding throughput gains are 13x, 18x, and 17x. Client-side microbenchmarks with simulated network delay show speedups of 12x–20x (online masking), 34x–135x (precomputed), and 38x–157x (verification only).
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
The source did not state who this applies to in practice.
An early-stage prototype implementation of a novel cryptographic protocol for LLM inference, demonstrating feasibility and throughput gains but lacking peer review, real-world deployment validation, or comparison against established privacy-preserving inference methods.
As stated by the source record.
Quoted from the source exactly as published.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
Open-source large language models (LLMs) are increasingly competitive with closed-source models while offering transparency and the ability to run inference without exposing user inputs to a service provider. However, running large-scale models locally requires substantial computational resources. In practice, users may still resort to a third-party provider, giving rise to privacy and correctness concerns. Existing solutions that address these problems often impose substantial server overhead or introduce additional trust assumptions. In this paper, we present Maverick, a novel approach to private and verifiable LLM inference based on a protocol for delegating matrix-vector multiplication, a dominant operation in LLMs. At its core, Maverick provides, to our knowledge, the first information-theoretically sound verification protocol for matrix-vector multiplication delegation with transparent preprocessing, efficient (batch) verification, and virtually no server overhead. We combine this verification primitive with LPN-based pseudorandom masking to provide input privacy. We implement our matrix-vector delegation primitive and use it to build an end-to-end prototype of Maverick, which we evaluate on Qwen3-4B by measuring throughput in tokens per second. We evaluate client configurations with 1-8 threads. With one client thread and a CPU server using up to 128 threads, Maverick achieves throughput gains over local inference of up to 17x when privacy masks are generated online, 45x when they are precomputed, and 44x when only verification is required. With four client threads, the corresponding gains are 13x, 18x, and 17x. When server computation is no longer the bottleneck, client-side microbenchmarks with simulated network delay show speedups of 12x-20x, 34x-135x, and 38x-157x.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.