Mission.

We're making high-assurance a core part of AI safety.

AI is improving faster than our ability to understand it. Most safety efforts (evals, mech interp, learning dynamics) focus on understanding how systems behave in known cases.

We view AI through a cryptographic lens to determine which safety properties can be guaranteed against capable, adaptive adversaries, and which cannot. We're building on years of research at the intersection of cryptography and AI safety.

We're creating an independent nonprofit to work in the open and optimize for public benefit. Stay tuned for updates.