Version 2 (April 2026) — refocused core to three primitives (PolicyDocument, ConsentRecord, AdherenceEvent). Extensions (tiered escalation, category preferences, regulatory context, audit projection) moved to separate protocol extensions under development in the GitHub repository. Added reference Python implementation with 35 tests, a reproducible micro-benchmark, and a working end-to-end demo with Google Gemini. See §4.3 and §4.4 of the PDF. Companion repository: https://github.com/ravikiran438/agent-consent-protocol --- As autonomous AI agents increasingly call other agents to complete tasks, a structural accountability gap has emerged: agents accept the terms of service of the services they call without any mechanism to prove they understood or subsequently honoured those terms. Authentication protocols establish who may call what. They say nothing about under what conditions a permitted call may be made, conditions that change as policies evolve. We formalise this as the distinction between proof of acceptance (a timestamped acknowledgement) and proof of adherence (a clause-level reasoning record produced at every action). We propose three primitives — PolicyDocument, ConsentRecord, and AdherenceEvent — that together constitute a versioned, append-only consent model for agent-to-agent communication. We instantiate the model as concrete extensions to two major agent protocols:the Agent2Agent (A2A) protocol and the Model Context Protocol (MCP). The model provides technical infrastructure designed to support the consent obligations of the EU AI Act ahead of the August 2, 2026 enforcement date for high-risk AI systems (Arts. 14, 50) and fills an identified gap in the FINOS AI Governance Framework.
Ravi Kiran Kadaboina (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: