Hi, I'm Prakrat!
I aim to work on important and interesting problems by doing good science. I'm currently focused on helping make AI go well for humanity and mitigating catastrophic risks from advanced AI.
I'm currently a MATS 10.0 scholar under Megan Kinniment (METR), working on AI evaluations for risk-relevant capabilities. I also study Math and Computer Science at UC Berkeley. On campus, I lead the Berkeley AI Safety Initiative, teach the Intro to AI Safety class, and am a member of Machine Learning @ Berkeley.
Previously, I did the Pivotal Fellowship under Jérémy Scheurer (Apollo Research), where I built a benchmark to measure circumvention propensity in coding agents. And before that, I studied self-preservation behaviors in LLMs under Robert McCarthy (UCL). I also organized IBAR 2025: a retreat for university AI Safety groups on the East Coast. In a previous life, I interned as a Software Developer at Amazon, helped teach the biggest class at Berkeley, and played competitive table tennis.
I enjoy reading, hiking, teaching, learning new skills, pizza, ultimate frisbee, table tennis, spikeball, board games, and house music. Have a low bar for reaching out!