Research Intern - Self-Improving & Secure AI Agents - Microsoft Research
Core
Develop and evaluate self-improving AI agents that use post-training experiments to iteratively improve target models via synthetic data, and design security mechanisms for AI agent systems.
Role type
Research Intern (Self-Improving & Secure AI Agents)
Builds
Self-improving AI agents with closed-loop optimization and security mechanisms
Domain
Artificial Intelligence, Machine Learning, Cybersecurity
Deliverable
research
Required skills
Python programming, PyTorch, mathematical modeling, software engineering, experimental design, data synthesis, capability diagnosis, security mechanism design
Preferred skills
Open-source contributions, competitive programming awards, publications at top-tier ML conferences
Technologies
PyTorch, Python
Responsibilities
Develop methods for task decomposition and capability diagnosis; build capability-targeted data synthesis methods; implement closed-loop optimization for self-improvement; develop least-privilege authorization and approval mechanisms for sensitive actions
Seniority
Intern