About the role#
This internship focuses on AI alignment and safety within frontier systems. You will join a team that connects evaluation, automated red teaming, and post-training to study emerging risks like deception, sandbagging, and unsafe autonomous behavior. Interns work on open research problems and collaborate with researchers and engineers to turn experimental findings into scientific insights, publications, and improvements to deployed systems.
What you'll need#
- Current enrollment in a Doctorate Degree program in Computer Science, Machine Learning, Artificial Intelligence, or a related technical field.
- Demonstrated research background in machine learning and foundation models.
- Experience conducting empirical research with large language models.
- Familiarity with AI alignment, frontier evaluations, red teaming, or post-training.
- Strong programming, experimentation, and quantitative analysis skills.
- Experience with post-training methods such as reinforcement learning, supervised fine-tuning, or preference-based methods.
- Interest in studying emerging behaviors like reward hacking and agentic misalignment.
- Ability to submit a cover letter, research samples, and a minimum of two reference letters.
What you'll do#
- Own a scoped research project in AI alignment, frontier evaluations, red teaming, or safety post-training.
- Design and run experiments on frontier models and agentic systems.
- Build new evaluations and automated red-teaming approaches to identify emerging safety risks.
- Develop or evaluate post-training techniques to improve model behavior.
- Collaborate with the team to produce research insights and contribute to the community.
- Present your findings throughout the internship.
Location & details#
- This is a full-time, paid, on-site internship.
- You must be physically located at the manager’s Microsoft worksite in either Redmond, Washington, or Mountain View, California, for the duration of the term.
- Applications are accepted on a rolling basis.


