Cryptography promises that no adversary can cheat the system. We are that adversary. False Witness turns AI loose on the cryptography and security the world runs on — the proofs, protocols, and primitives — to break what's deployed, to offer the breaking as a service, and to prove the next generation of code can't be broken at all.
Finding the bugs in cryptography and security that others miss — with AI, in both directions: breaking what's deployed, proving what's next.
We point AI at the open-source cryptography everyone quietly trusts — proof systems, MPC, threshold signatures, post-quantum schemes — and find the bugs before adversaries do. Reported, fixed, published, in the open.
The same machinery, pointed at your code. An orchestra of AI agents reads your implementation the way an auditor does — reasoning about protocol invariants, not just syntax — and reports soundness gaps and implementation bugs continuously.
Bugs are cheapest to kill before they exist. We use AI proof assistants to formalize cryptographic proofs in Lean4 — machine-checked theorems that become the spec the implementation is held to. clean is our verification DSL for exactly that.