Behavior-Bound Signatures: Policy Compliance via Zero-Knowledge Soundness

When an AI agent holds a private key, every existing signature scheme gives it unrestricted power to sign any transaction — including transactions its principal never authorized. Multi-signature, smart-contract guards, and hardware enclaves are all soft constraints: they live outside the signature's mathematical structure and can be bypassed. We present behavior-bound signatures (BBS), the first signature scheme in which even the private-key holder cannot sign messages that violate a self-imposed behavioral policy. The core mechanism embeds a zero-knowledge proof inside each signature, attesting that the signed action satisfies a policy constraint δ(x) < ε committed at key generation. If the action violates the policy, no valid signature exists — not because the signer refuses, but because the ZK constraint system is unsatisfiable. We formalize policy-soundness (PS-CMA), a security model strictly stronger than EUF-CMA, covering both identity forgery and compliance forgery. We prove PS-CMA security under standard assumptions (Pedersen binding, Poseidon collision resistance, Bulletproofs/PLONK soundness) and show that the compliance filter rejects all non-compliant actions even under full Byzantine control (f ≤ n−1) — decoupling compliance safety from honest-quorum requirements entirely. We demonstrate the framework with autonomous AI agents, providing the first cryptographic guarantee that a compromised or malicious agent cannot exceed its authorized behavioral envelope.