Newest papers

Proof-Carrying Contracts: Commitments AI Agents Cannot Break

Carlo Ciliberto, Alessandro Rudi, Enrico Tassi · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28484 · cited by 0 works (OpenAlex) · open-access PDF

French

Les agents d'intelligence artificielle agissent de plus en plus au nom des personnes ou d'autres agents, achètent des services, échangent des informations et déléguent le travail. Ce qu'un agent doit respecter est écrit dans des instructions, des politiques ou des accords, mais le texte seul ne peut le faire respecter. Il peut faire des erreurs, être redirigé par ce qu'il lit, ou poursuivre des intérêts différents. Pour qu'une économie de ces agents puisse fonctionner, les parties ont besoin d'accords qu'elles peuvent compter sur sans inspecter ou faire confiance aux agents les uns des autres. Nous introduisons un système d'exécution de modèle appelé Airlock, un cadre permettant de rendre les contraintes convenues exécutoires tout au long des interactions d'un agent avec un système.

Machine translation (Meta M2M100); the English original follows.

Abstract

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. 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. 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.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Permission, Computer security, Adversarial system, Computer science, Purchasing, Foundation (evidence)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange