AgenTEE: Trusted Execution Environments for Confidential LLM Agents on Edge Devices
arXiv·medium signal
AgenTEE addresses the security gap in running LLM agents on edge devices by leveraging Trusted Execution Environments (TEEs). The paper identifies that LLM agents create a broader attack surface than traditional apps due to non-deterministic model behavior and third-party service integration. The framework provides confidential execution of complex agent pipelines on edge hardware — relevant as more agent workloads move from cloud to device for latency and privacy.