Our work Bileve is accepted by NeurIPS 2024. 🎉 We propose a bi-level signature scheme to safeguard LLM-generated texts against both forgery and evasion attacks.