AI Agent Security Research Engineer

Who We Are At OKX, we believe that the future will be reshaped by crypto, and ultimately contribute to every individual's freedom. OKX is a leading crypto exchange, and the developer of OKX Wallet, giving millions access to crypto trading and decentralized crypto applications (dApps). OKX is also a trusted brand by hundreds of large institutions seeking access to crypto markets.

We are safe and reliable, backed by our Proof of Reserves. Across our multiple offices globally, we are united by our core principles:
We Before Me, Do the Right Thing, and Get Things Done. These shared values drive our culture, shape our processes, and foster a friendly, rewarding, and diverse environment for every OK-er. OKX is part of OKG, a group that brings the value of Blockchain to users around the world, through our leading products OKX, OKX Wallet, OKLink and more.

Design and implement a multi-agent collaborative code auditing system covering vulnerability detection, malicious code identification, and sensitive information leakage scenarios; lead the role decomposition of Planners/Executors/Critics, tool invocation chains, and cross-agent state synchronization mechanism design.

Integrate RAG, Chain-of-Thought, Reflection, and other technologies into security audit agents. Continuously optimize detection accuracy and recall rates while establishing a quantifiable evaluation and iteration framework.

Deeply integrate with Dev Sec Ops  workflows. Develop plugins for mainstream pipelines like Git Lab CI/CD, Tekton, and Jenkins to achieve “audit-on-commit.”

Responsible for constructing a security protection framework for large language model applications, covering three dimensions: input layer (prompt injection, jailbreak detection), output layer (sensitive information leakage, compliance auditing), and runtime (tool invocation sandboxing, anomaly behavior circuit breaking).

Develop Agent workflows for automated alert classification, contextual correlation, and false positive filtering. Integrate RAG-driven threat intelligence retrieval to generate automated analysis conclusions, supporting SOAR platform integration.

Design human-machine collaboration intervention mechanisms and Agent behavior audit systems to ensure observability, traceability, and intervenability of Agent actions in production environments, adhering to industry standards like the OWASP Top 10 Risks for LLMs.

Construct a highly available, scalable Agent service architecture supporting large-scale concurrent scanning task scheduling and fault tolerance. Oversee standardized API output for detection capabilities, building closed-loop systems for rule management, result visualization, and false positive feedback.

Development

Experience:

3+ years of backend development experience, proficient in at least one of Python/Go/Java, with a solid engineering foundation. Agent Implementation & Security:
Hands-on experience deploying LLM Agents (not just demos), capable of detailing engineering challenges such as Agent architecture design, hallucination handling, and tool invocation fault tolerance;
Hands-on experience with AI security, understanding risks like prompt injection, jail breaking, malicious agent injection, and tool misuse, with implementable defense strategies. Framework Proficiency:
Familiarity with at least one agent framework (Lang Chain, Llama Index, Auto Gen, CrewAI, or Lang Graph), with production project experience. Engineering Capabilities:
Proficient in Docker and Kubernetes, with expertise in microservices architecture design and deployment.

Security Tool

Experience:

Experience with SAST/SCA tools, or deep usage of code auditing tools like CodeQL, Semgrep, or Sonar Qube.

Model Fine-Tuning:
Experience with LLM fine-tuning (SFT, LoRA), or familiarity with local deployment and optimization of models like Llama 3, Qwen, or Deep Seek. Bonus points for security-domain fine-tuning experience, such as training and evaluating security detection models for malicious prompt…