Hey, I'm Ram

I build the safeguards that monitor and control advanced AI agents

I'm an AI control and monitoring researcher and a Fellow at Redwood Research, where I design and red-team the safeguards that catch a misaligned AI agent before it causes harm, through control evaluations, monitors, and automated red-teaming pipelines. I was second author of LinuxArena, a control arena I ran all the evaluations for, now used for risk evaluations at Anthropic and the EU AI Office. Earlier I founded and exited an AI startup, then realized AI risk is a huge issue. I hold a B.S. in Artificial Intelligence from Carnegie Mellon.