SEAL: Secure and Efficient Adaptive Layering for On-Device Language Models with TEE
Journal
Proceedings - 2025 IEEE 11th International Conference on Edge Computing and Scalable Cloud, EdgeCom 2025
Start Page
138
End Page
143
ISBN (of the container)
979-833158779-6
Date Issued
2026-01-22
Author(s)
Abstract
The deployment of Small Language Models (SLMs) on edge devices presents a fundamental trade-off between performance and security: executing in the untrusted Rich Execution Environment (REE) risks model theft, while running entirely within a Trusted Execution Environment (TEE) incurs significant overhead. Existing approaches often sacrifice one for the other. To address this, we propose SEAL (Secure and Efficient Adaptive Layering), a framework that enables informed, quantitative trade-offs between security and efficiency for on-device inference. SEAL introduces two key components: Confidential Layer Analysis (CLA), which quantitatively assesses the confidentiality of each model layer, and the Layer Importance-Guided Adaptive Partition (LIAP) algorithm, which maps the most sensitive, low-overhead layers into the TEE based on device constraints, while retaining others in the REE to preserve performance. To evaluate security, we develop the Model Parameter Reconstruction Success Rate (MPRSR), a metric that measures behavioral, parametric, and functional similarity under model extraction attacks. Experiments using the INT4-quantized Qwen3-0.6B model on WikiQA demonstrate that SEAL reduces model theft risk by 65.8%—with only a 22% increase in latency and memory—by protecting a single critical layer. When securing the top five layers, SEAL reduces MPRSR to 7.9%, while reducing time and energy consumption by roughly 50% compared to full-TEE deployment. SEAL reframes security as an optimization problem, enabling practical and trustworthy secure inference for edge AI.
Event(s)
11th IEEE International Conference on Edge Computing and Scalable Cloud, EdgeCom 2025
Subjects
Edge Computing
Energy Efficiency
Model Stealing
On-Device AI
Security and Privacy
Small Language Models
Trusted Execution Environment (TEE)
Publisher
IEEE
Type
conference paper