About Me
Who I am
Section titled “Who I am”Hello, and welcome to XiaozaoSec.
I am a cybersecurity engineer working across red-team/blue-team operations, penetration testing, code auditing and AI security. The field moves quickly, so I keep notes while I learn and practise — XiaozaoSec is those notes, tidied up and published.
“Sec” stands for information security. “Xiaozao” is the name of my hometown; using it here is a small personal indulgence.
Articles on this site are written and organised by me, and I credit external sources where I rely on them.
What this site is
Section titled “What this site is”XiaozaoSec is an AI security knowledge base for practitioners, focused on the defensive side.
The attack surface of AI systems has expanded very quickly, but most available material stops at definitions and leaves out what to actually do. This site tries to cover the second half: for each class of risk, first why it happens, then the defensive actions that really work. Content is organised by topic as a coherent body of work rather than a pile of one-off posts.
Everything is published in both Chinese and English, so readers can work in whichever language they prefer.
Research focus
Section titled “Research focus”My work covers defensive security and red-team research across these areas:
- LLM Security: prompt injection, jailbreak attacks and mitigations, and the vulnerability mechanics and defences behind the OWASP LLM Top 10.
- Agent Security: risks in AI agents, including the OWASP Agent Top 10 and A2A protocol security.
- MCP Security: practical MCP usage, plus the mechanics, risk and mitigation of the OWASP MCP Top 10.
- Skill Security: practical agent-skill usage, plus the mechanics, risk and mitigation of the OWASP Skill Top 10.
- AI-Enhanced Cybersecurity: putting AI to work on the defensive side — assisted vulnerability discovery, code audit and security operations.
- Robotics Security: security for humanoid robots and embodied AI, covering communication links, sensor spoofing and actuation permissions.
- AI Data Security: privacy and integrity risks across training corpora, vector stores and the inference path.
How the content is organised
Section titled “How the content is organised”- Every category opens with an overview explaining what the topic covers and the order worth reading it in.
- Articles carry a publication date, and a “last updated” date whenever the content is revised.
- If you spot an error, or disagree with a conclusion, email me and I will check it and revise the article.
Get in touch
Section titled “Get in touch”- Email: [email protected] — the most reliable way to reach me, and I will reply as soon as I can
- WeChat: xiaozaosec — for technical discussion; I run a study group
- Other platforms: see the Contact page
I will keep publishing as the field develops, and I am always glad to compare notes. Wish you good health, peace and rewarding learning experiences.