Skip to content
#small language model Book Open access

TWIST: Private Tenant-Dedicated LLM Serving via Joint Model-Token Transformation

Nov 2026 · Proceedings of the 2026 ACM SIGSAC Conference on Computer and Communications Security · 0 citations · 16 references

TL;DR

This work presents TWIST, a practical system for private open-weight LLM serving on dedicated cloud instances that applies a shared, structure-preserving transformation to both the model weights and input tokens, extending protection beyond data-only obfuscation and addressing the traditional utility bottleneck.

Abstract

Organizations increasingly host open-weight large language models (LLMs) on dedicated cloud instances to avoid the opaque data control imposed by closed-source APIs and the burden of costly local infrastructure. However, this deployment paradigm inadvertently exposes sensitive tenant prompts and outputs to service providers. Existing protection routes do not fit this setting well in practice: cryptographic inference incurs prohibitive overheads; Trusted Execution Environments (TEEs) rely on specialized hardware that is difficult to deploy consistently across heterogeneous legacy accelerators; and traditional data-only obfuscation methods, such as token replacement, require substantial perturbation to achieve meaningful privacy, often causing severe utility degradation. To address this gap, we present TWIST, a practical system for private open-weight LLM serving on dedicated cloud instances. Rather than perturbing input tokens alone, TWIST applies a shared, structure-preserving transformation to both the model weights and input tokens. This coupled design extends protection beyond data-only obfuscation and addresses the traditional utility bottleneck. We analyze leakage to an honest-but-curious provider using Rényi-metric differential privacy (RmDP), and show a smaller privacy budget than a data-only baseline with the same online mechanism. Practically, we implement TWIST for modern dense and Mixture-of-Experts (MoE) architectures, including attention, FFN, and MoE routing modules, without disrupting standard serving pipelines. Evaluations on the 671B-parameter DeepSeek-V3.1-Terminus model show that TWIST reduces token recovery rate to under 5% against advanced inversion attacks, while incurring only 0.0% - 3.5% accuracy loss and <10% inference speed loss, offering a practical solution for private tenant-dedicated LLM serving.

Read PDF

Similar papers

#small language model Dataset Open access Oct 2026

Socratic guiding questions in synthetic arithmetic data: matched LoRA runs (revision v2)

Supporting data, adapters, predictions and code for the article *Low-Cost LoRA Fine-Tuning of Small Language Models for Multi-Step Arithmetic Reasoning* by Jake O'Grady, Asena Isik Gürhan, Chee Fong Ting and Effirul Ramlan (University of Galway). We generated 20,000 GSM8K-derived arithmetic problems with step-by-step s...

O'Grady, Jake, Gürhan, Asena Isik, Chee, Fong Ting et al. · 465 citations
#computer vision Open access Jun 2016

Software Development in Startup Companies: The Greenfield Startup Model

The results are packaged in the Greenfield Startup Model (GSM), which explains the priority of startups to release the product as quickly as possible, and the need to shorten time-to-market, by speeding up the development through low-precision engineering activities.

Carmine Giardino, Nicolò Paternoster, M. Unterkalmsteiner et al. · 178 citations · ⚡14
#computer vision Open access Oct 2016

Software Startups - A Research Agenda

Software startup companies develop innovative, software-intensive products within limited timeframes and with few resources, searching for sustainable and scalable business models.

M. Unterkalmsteiner, P. Abrahamsson, Xiaofeng Wang et al. · 157 citations · ⚡17
#machine learning Review Open access Oct 2016

“Failures” to be celebrated: an analysis of major pivots of software startups

This study conducts a case survey study based on the secondary data of the major pivots happened in 49 software startups, and demonstrates that customer need pivot is the most common among all pivot types.

Sohaib Shahid Bajwa, Xiaofeng Wang, Anh Nguyen-Duc et al. · 127 citations · ⚡15
#computer vision Review Open access May 2015

A survey study on major technical barriers affecting the decision to adopt cloud services

The comparison of adopter and non-adopter sample reveals three potential adoption inhibitor, security, data privacy, and portability, which underlines the importance of the technical and security perspectives for research investigating the adoption of technology.

Nattakarn Phaphoom, Xiaofeng Wang, S. Samuel et al. · 111 citations · ⚡8
#large language models Review Open access Nov 2020

Status of the WHO recommended timing and frequency of antenatal care visits in Northern Bangladesh

Objective There is dearth of information on the timeliness of antenatal care (ANC) uptake. This study aimed to determine the timely ANC uptake by a medically trained provider (MTP) as per the World Health Organization (WHO) recommendations and the country guideline. Methods Cross-sectional survey was done with 2,731 wo...

B. Sarker, Musfikur Rahman, T. Rahman et al. · 89 citations · ⚡7

Related blog posts

Microsoft Research Blog Nov 16, 2026

TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient

AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.