跳转至

携带证明合约 (PC2)：AI 智能体无法违背的硬性承诺

Proof-Carrying Contracts: Commitments AI Agents Cannot Break

arXiv: arXiv:2610.28484 [cs.AI]
Authors: Carlo Ciliberto, Alessandro Rudi, Enrico Tassi
Submitted: October 20, 2026
Subjects: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Logic in Computer Science (cs.LO); Multiagent Systems (cs.MA); Systems and Control (eess.SY)

文章背景与核心概要

随着自主 AI 智能体 (AI Agent) 逐渐深入人类社会的方方面面，它们开始代表用户自主购买服务、交换核心信息并委派协作任务。然而，目前业界仅依靠自然语言指令或静态文本策略来约束智能体行为，存在严重的可靠性隐患——智能体既可能产生逻辑失误或目标偏离，也容易沦陷于提示词注入 (Prompt Injection) 攻击。针对这一协作信任瓶颈，研究团队开创性地提出了“携带证明合约 (Proof-Carrying Contracts, PC2)”体系，将模糊的行为准则升华为机器可严格检验的数学约束。任何智能体在执行关键动作前，都必须出具一份完备的数学证明，并经由名为“气闸 (Airlock)”的系统核验放行，从而在赋予智能体充分策略自由度的同时，筑牢了坚不可摧的底层安全防线。

📌 内容概要

📌 Summary

随着 AI 智能体 (AI Agent) 越来越普遍地代表用户自主行动——从采购各类服务、交换关键数据到委托分派工作——单纯依靠文本形式的指令或规则策略，根本无法确保其行为绝对合规。智能体随时可能产生失误、受到提示词注入 (Prompt Injection) 攻击而被恶意诱导，甚至转而追求与用户冲突的其他目标。

As AI agents increasingly act autonomously on behalf of users—purchasing services, exchanging information, and delegating work—relying solely on textual instructions or policies is insufficient to guarantee compliance. Agents can make mistakes, be manipulated by prompt injection, or pursue conflicting interests.

为了彻底破局，论文作者提出了携带证明合约 (Proof-Carrying Contracts, PC2)：这是一种全新的系统框架，能够在智能体与外部系统的交互全生命周期中，以严谨的数学手段强制执行既定约束。在执行任何请求的操作之前，智能体必须先提供一份机器可自动核验的数学证明，以确保该操作在任何可能的情境下均能严格维持既定约束。名为气闸 (Airlock) 的底层执行系统会在放行该操作前对证明进行快速核查；这种机制既能让智能体灵活地自适应调整求解策略，又能为规则底线的绝对守序提供确定性保障。

To solve this, the authors introduce Proof-Carrying Contracts (PC2): a novel framework that makes agreed constraints mathematically enforceable throughout an agent's interactions. Before executing any requested operation, the agent must supply a machine-checkable proof confirming that the operation preserves all agreed-upon constraints. An execution system called the Airlock verifies this proof before granting access, freeing the agent to adapt its strategy while ensuring absolute constraint adherence.

📋 论文摘要

📋 Abstract

AI 智能体正日益代表人类或其他智能体独立行动，参与购买服务、互通信息以及委派工作。智能体必须恪守的底线通常写在指令、规则策略或协议条款中，但光凭冰冷的文字本身并不能强制其完全顺从。智能体可能在执行中犯错、被读取的恶意输入篡改意图，或是追求与委托方相背离的利益。若想让这类智能体构成的协作经济生态真正运转起来，各参与方亟需一种坚实可靠的契约形式，即便无需费力审查对方智能体的内部实现或盲目托付信任，也能安心建立协作。

AI agents increasingly act on behalf of people or other agents, purchasing services, exchanging information and delegating work. What an agent must respect is written in instructions, policies or agreements, but text alone cannot make it comply. It may make mistakes, be redirected by what it reads, or pursue different interests. For an economy of such agents to function, parties need agreements they can rely on without inspecting or trusting one another's agents.

我们在此提出携带证明合约 (Proof-Carrying Contracts, PC2)，这是一个能够让约定的约束条件在智能体与系统交互全过程中具备强制执行效力的架构。在任何被请求的操作得以运行之前，智能体必须主动递交一份机器可自动检验的数学证明，证明提议的操作在其行为模型所允许的每一种可能情况下，均能严格保留既定约束。名为气闸 (Airlock) 的执行系统将对该证明展开严格校验，并以此掌控对底层操作权限的开放与拦截。

We introduce Proof-Carrying Contracts (PC2), a framework for making agreed constraints enforceable throughout an agent's interactions with a system. Before a requested operation can run, the agent must supply a machine-checkable proof that it preserves the agreed constraints in every case allowed by its model. An execution system called the Airlock checks the proof and controls access to the operation.

在这一体系中，协作各方只需专注明确需要保留的安全条件，而智能体自身则负责提供证明其决策合法性所需的推理逻辑，同时享有自如调整策略的充分空间。这避免了为系统费力设计繁复的静态权限策略，并逐项证明其允许的每个决策都安全可靠的工程负担。我们在理论上证明，即便面对恶意对抗的智能体，这些约束在运行周期的每一个瞬间都能严格成立，并厘清了何时能将该数学安全性平移至现实世界的实际安全。此外，可复用的能力模块为合约的组装提供了现成的工具库与理论保障。PC2 框架为未来的智能体经济打下了可信基石，使达成的协议既能有效保护委派工作的发起方，也能切实保障具体承担任务的智能体本身。

The parties specify the conditions to preserve, while the agent supplies the reasoning needed to justify its decisions and remains free to adapt its strategy. This avoids having to design a separate permission policy and prove that every decision it permits is safe. We prove that the constraints hold at every instant during execution, even for an adversarial agent, and identify when this implies safety in the world. Reusable capability modules provide tools and guarantees from which contracts can be constructed. PC2 provides a foundation for an agent economy whose agreements protect both those delegating work and those carrying it out.

🔗 全文与相关资源

🔗 Full-Text & Resources