01
Contract semantics and intent
How can a formal contract capture what each principal intends? We study contract languages, interactive formalization and the conditions under which contractual guarantees compose.
An open protocol for AI agents
Proof Carrying Contracts
A commitment layer for the agent economy.
The agent protocol stack is taking shape around MCP for tools and data, A2A for agent communication, and AP2 for payment authorisation. PC2 is developing a general commitment layer for this stack: contracts that govern actions, preserve obligations through delegation, and make completion checkable between principals whose interests may conflict.
The mission
Enforceable agreements between distrusting principals
Agents act on behalf of principals: the people or organisations whose goals and resources they represent. When agents exchange services or delegate work, their principals may have conflicting interests and need commitments they can verify without relying on the counterparty's agent to comply voluntarily.
The central idea
Agents agree, act and delegate under a contract
The parties define the task and its conditions in a shared contract. The execution mechanism checks the required evidence before permitting an action, including an action that delegates part of the task to another agent.
01Establish a shared contract
The participating agents agree on the task, resources, constraints and completion conditions. Their principals approve the contract and the execution environment.
02Check before execution
A proposed action is checked against the contract, current state and declared tool effects. An accepted proof permits the governed tool call.
03Delegate within the agreement
A subcontract assigns a defined subtask and its authority to another agent. Its commitments must remain consistent with the parent contract and existing delegations.
The research programme
The research questions behind PC2
The programme studies whether useful requirements can be formalized faithfully, whether agents can act under them at practical cost, and what protection the resulting system provides.
02
Verified action and delegation
How can agents produce the evidence required for useful work? We develop proof methods, verified tools and execution infrastructure that preserve contractual constraints through action, delegation and settlement.
03
Adversarial evaluation
What protection does PC2 provide, and at what cost? We evaluate contract fidelity, prevented violations, useful task completion and computational overhead in benign and adversarial multi-principal interactions.
PC2 foundations paper
Proof-Carrying Contracts: Commitments AI Agents Cannot Break
The paper develops the mathematical model of contracts, execution and delegation, with proofs under stated assumptions.
The mechanization and code accompanying the PC2 paper are coming soon in the project repository. A demonstration will follow, with instructions for running an initial example and examining its contract and execution.
Demonstration · UpcomingGitHub repository
Research consortium
Research for the economy of AI agents
PC2 brings together research groups at SDA Bocconi, UCL, AI4I and Inria, connecting machine learning, formal verification, programming languages, agent systems and the economics of cooperation.
This combination connects the protocol's formal guarantees with the incentives and conditions that make exchanges between independent parties viable. The shared objective is to develop commitments that agents can establish, rely on and preserve through execution.
Get involved
Build the science and infrastructure of PC2
Research and engineering appointments span contract theory, intent contractualization, agent learning, execution infrastructure and interactions between distinct principals. Joining the programme provides scope to develop an identifiable contribution to its methods and public artifacts.
Researchers and developers can also take part through scientific discussion, challenging use cases and, as implementations are released, testing and code contributions.