What we work on

Hypervisor-Zero is an independent AI safety research project focused on building tools to test, monitor, and control autonomous AI agents. It works on agent runtime security, adversarial evaluation, and AI-powered cyber defense, with an emphasis on verifying whether safety policies are actually enforced.

Who's doing it?

The project is led by me I'm a software engineering student and solo independent researcher

Where it's up to?

I moved Hypervisor-Zero from idea stage to a public demo, codebase, a working prototype, and a researcher dashboard Its core runtime security and agent-control components are under development, alongside a system for monitoring tool calls, enforcing policies, detecting adversarial behavior, and recording evidence of whether protected actions were blocked. The project is also developing ControlBench, its evaluation dataset, and a Tinker-based fine-tuning pipeline.