微信安全-大模型安全工程师
Core
Develop and optimize content safety risk control strategies for large language models (LLMs) and agents, focusing on risk identification, attack defense, and automated red team evaluation.
Role type
Senior LLM Security Engineer (Content Safety & Red Teaming)
Builds
Production safety guardrails, automated red team testing frameworks, and model safety evaluation tools.
Domain
Artificial Intelligence / Large Language Model Security
Deliverable
production ML models
Required skills
LLM architecture (SFT, DPO, RAG, Agent), Prompt Injection defense, Adversarial attack mitigation, Red teaming methodologies, Data analysis, Python engineering
Preferred skills
Model fine-tuning experience, Model watermarking & tracing, Compliance policy interpretation
Technologies
LLM frameworks, Python, Data analysis tools
Responsibilities
Design and tune risk identification rules for LLM content safety; Analyze and defend against attacks like prompt injection and data poisoning; Build automated red team evaluation systems and test cases; Monitor and iterate safety strategies from experiment to full rollout; Implement security checks for Agent call chains and model tracing; Track emerging security technologies and compliance regulations.