About

My name is Dave Banerjee, and I am an AI policy fellow at the Institute for AI Policy and Strategy, where I research AI security, compute governance, and international verification. Right now, I’m writing papers on AI integrity, the problem of ensuring that frontier AI systems are not tampered with (e.g. preventing nation state actors from backdooring American models).

Previously, I did the following stuff:

I was a summer fellow at the Centre for the Governance of AI, where I threat modeled risks from secretly loyal AIs and investigated whether the US or China could robustly hide data centers (AKA Dark Compute).
I was a participant in ARENA 5.0, a 5-week ML alignment bootcamp. In my ARENA capstone project, I investigated whether self-perceived superintelligent LLMs (LLMs convinced that they are superintelligent via prompting) exhibit misalignment.
I was a research fellow in the SPAR program, where I forecasted how much it costs to hack an Nvidia H100 (i.e. how much money, time, equipment, and expertise is required to leak various cryptographic keys off the chip).
I was a security engineer at a boutique hedge fund, where I built their security infrastructure from the ground up and led all security operations.
I graduated from Columbia University in December 2024, where I studied computer science with a focus on cryptography, reverse engineering, and ML.

My academic interests lie in AI security & governance, hardware & software security, (meta)ethics, physics, and cognitive science. I enjoy thinking about what a post-AGI world will look like and how we can make it as good as possible.

If you would like to receive emails whenever I post: Subscribe Here!

I have signed no contracts or agreements whose existence I cannot mention.

I hope you enjoy exploring my site :)

— Dave