Open Source Projects

A lightweight built-in security plugin for OpenClaw that enforces defense-in-depth runtime protection across the full LLM agent lifecycle — from initialization to execution.

The first multi-level safety evaluation platform for OpenClaw-style AI agents, jointly released by Tsinghua University, Peking University, Zhejiang University, Nanjing University, Hangzhou Dianzi University, and Ant Group. Provides a three-tier (L1/L2/L3) evaluation framework over 7 risk dimensions.

A full-stack security operating system for trustworthy AI agent deployment, unifying onboarding, secure reasoning, and trusted execution through a heterogeneous defense-in-depth architecture.

A PyTorch-based open-source benchmark library implementing 11 state-of-the-art deep learning website fingerprinting attacks on a unified, reproducible framework with multi-dataset support.