Undergraduate Researcher — AI Safety
Study when and why models behave in unintended ways: alignment, evaluations, red-teaming, and robustness of large language models and agents. Ideal if you have some ML background and strong Python skills.
Apply →ASTRA — AI Security, Trust, Resilience, and Assurance — is a small research group at UC Berkeley. We're looking for motivated undergraduates to conduct research in AI safety, AI security, and AI for security.
AI systems are being deployed faster than we can understand their failure modes. ASTRA studies how to make them safe, secure, and reliable — and how to turn them into tools that defend, not just systems that need defending.
You work on open problems with a path to publication — designing experiments, building systems, and writing up results.
Small group, close guidance. You meet regularly with experienced researchers who invest in your growth.
You take a research question from idea to result — driving it end to end and presenting your progress to the group.
All positions are for current UC Berkeley undergraduates, with flexible weekly hours during the semester.
Study when and why models behave in unintended ways: alignment, evaluations, red-teaming, and robustness of large language models and agents. Ideal if you have some ML background and strong Python skills.
Apply →Attack and defend AI systems themselves: adversarial examples, jailbreaks, prompt injection, data poisoning, and securing agentic pipelines. A security mindset matters more than prior AI experience.
Apply →Use AI to strengthen software and systems security: LLM-assisted vulnerability discovery, program analysis, fuzzing, and automated threat detection. Systems or security coursework is a plus.
Apply →Not sure which area fits? Apply anyway and tell us what excites you — we'll help you find a project.
The form takes about 10 minutes. We read every application.