I build coordination and safety infrastructure for AI that acts in physical space. At Tethral I'm developing the layer between an AI's reasoning and its physical actuation: what an agent is authorized to do, how that authority is consented to and revoked, and how it fails safely. The company runs an MCP host with over 200,000 downloads and a digital-twin testing platform of more than 800 devices.

Before Tethral, I led AI and personal-safety product work in physical environments at Uber, where I shipped Uber Teens and co-led an enterprise design partnership with OpenAI. Earlier I worked on platform security at the US Department of Justice. My academic background is a Cornell PhD, with fellowships at MIT and Oxford, and my research runs across communication, information theory, and systems design. I've served as an invited safety tester for Anthropic.

The through-line in my work is writing the rules for new technical systems before the defaults get set by whoever ships first: autonomous vehicle policy in Massachusetts, generative AI policy at Uber, and now an open standard for embodied agents entering the home. I also publish open safety research; my most recent release, Shapestate, detects silent tampering in language models from their internal state, with pre-registered predictions validated across seven models.