AI & Software Security
织安实验室
Weaving intelligence into software security.
Weaving security into software.
The group
SecWeave Lab is Ting's research group, exploring the intersection of artificial intelligence and software security.
Our name reflects two complementary goals: weaving intelligence into software security, using AI to discover, understand, and repair vulnerabilities; and weaving security into software, integrating security throughout software design, development, and evolution.
Our mascot
Weavy is SecWeave's blue bird mascot, inspired by the craft of nest weaving. With SW on its chest and a thread in its beak, it mends a gap in code — a small expression of our goal to weave security into software.
Our focus
Building LLM agents that combine program analysis and execution to find and fix vulnerabilities in open-source software, checking each repair with exploit and functional tests, and making LLMs generate secure code in the first place.
Selected workTitanCA · Sifting the Noise · SeCuRepair · Vul4Py · SoR
Determining which releases, forks, and downstream packages a vulnerability affects, uncovering silent security fixes, and linking vulnerability reports to their fixing commits.
Securing AI coding agents and the skills, plugins, and MCP servers they install: how untrusted input hijacks an agent, how extension ecosystems are governed, and how agent systems behave under faults.
Selected workAttacker's Shell · Agent skill ecosystem study · AgentChaos
Our people