Undergraduate in Cyberspace Security at Sun Yat-sen University, working toward Fall 2027 M.S. programs in AI Security / AI Systems.
I make LLM and agent systems fail loudly instead of corrupting silently โ finding and fixing trust-boundary, data-fidelity, and scoring-correctness bugs in mainstream AI/ML open source.
Independent research on trust boundaries in MCP/agent frameworks (instruction trust-tier misclassification and prompt-cache amplification). Currently going through coordinated disclosure with the affected upstream project โ full write-up will be published once the fix lands.
Merged contributions across UK AI Safety Institute (inspect_ai scoring correctness), Trail of Bits (fickling pickle-security analyzer), NVIDIA (garak detector attribution), Microsoft (PyRIT payload fidelity), and the official MCP ecosystem (input-schema hardening, in review).
- Trust boundaries in LLM/agent frameworks (MCP, instruction routing, prompt-cache interactions)
- Correctness of AI evaluation & red-teaming toolchains
- Serialization security & static analysis
- Email: fei76161@gmail.com
- Application-relevant CV available on request