Neelasha Sudarshan

Neelasha Sudarshan

Undergraduate Student, Computer Science and Engineering

Santa Clara University

nsudarshan@scu.edu

Focus

I’m interested in working in AI safety because I want to create safe, useful technology that improves people’s lives. I have studied collusion in LLMs, AI Control, and interpretability. I’m looking for technical fellowships or research roles where I can help advance research in these domains. Feel free to contact me if you are interested in working together.

Projects

White-Box Steering Defenses for Safe AI

I tested white-box interventions to prompt injection attacks. I experimented with probe-gated steering methods to find the tradeoff between attack suppression and benign task performance.

Code · Write-up

Collusion and AI Control

I studied how collusion between AI agents can subvert AI control systems, through the mechanisms proposed in "When can we trust untrusted monitoring? A safety case sketch across collusion strategies" (Gardner-Challis et al.).

Code · Write-up