The Policy Stack
Four guardrail tiers for on-chain AI agents: probabilistic classifiers through Lean 4 theorem provers. Select an attack scenario to see which tier catches it — and which let it through.
⬢ loading artifact…
Appears in
The Compliance Kernel: Formal Verification for On-Chain AI Agents
Probabilistic classifiers miss 30–40 % of policy violations. Lean 4 theorem provers and SMT solvers make certain guardrail tiers mathematically certain — here's how the four-layer policy stack works.
8 min read ⬢ interactive