A significant governance proposal has emerged from the Gonka community, introducing a comprehensive framework for agent identity and delegation governance. The proposal, documented in GitHub Issue #922 and closed on March 22, 2026, presents the Agent Passport System (APS) as a solution to one of decentralized AI's most pressing challenges: how to grant AI agents secure, verifiable, and revocable permissions within distributed compute networks.
Cryptographic Agent Identity
The core innovation lies in providing each AI agent with an Ed25519 cryptographic identity, establishing a foundation for trust and accountability. The proposal outlines a sophisticated delegation system where humans can grant agents specific permissions with defined spend limits. These agents can then sub-delegate portions of their authority to other agents, with each delegation automatically narrowing the scope of permissions — ensuring authority monotonically decreases at each level of the chain.
Cascade Revocation
A critical security feature is cascade revocation: when any delegation in the chain is revoked, all downstream sub-delegations immediately become invalid. This prevents orphaned permissions and ensures complete control over access rights throughout the delegation hierarchy.
Three-Signature Audit Trail
The proposal introduces a three-signature policy chain that creates an immutable audit trail for every action. Each operation generates signatures for intent, policy evaluation, and execution receipt, providing unprecedented transparency and accountability in agent operations.
Impact on Gonka Infrastructure
For Gonka's decentralized inference network, this system represents a fundamental shift from API key-based access to cryptographically-verified agent authorization. When an agent requests inference services, the request carries a delegation proof that verifies the authorizing human, specifies which models and capabilities are accessible, defines spending limits, and provides signed receipts for billing purposes.
This transformation addresses a key limitation in current decentralized AI systems: the challenge of providing granular access control without complex API key management. Subnet operators gain fine-grained billing and access control capabilities while eliminating the overhead of managing individual API keys for each agent.
Technical Implementation
The Agent Passport System ships as both an MCP server with 61 tools and an npm package (agent-passport-system) backed by 866 tests. The proposal includes a gateway enforcement boundary that sits in front of the routing layer, validating delegation scope before forwarding requests to inference providers.
Interoperability was a key design consideration. The system has undergone cross-engine testing with three other governance protocols — AIP, Kanoniv, and Guardian — all utilizing Ed25519-based cryptography for consistency across different AI governance frameworks.
Resources
The community can access the full SDK at https://github.com/aeoess/agent-passport-system, while the technical details are documented in a research paper available at https://doi.org/10.5281/zenodo.18749779.
This proposal represents a mature approach to agent governance, addressing security, accountability, and scalability concerns that have hindered broader adoption of decentralized AI systems. The community's acceptance of this proposal signals Gonka's evolution toward a more secure and professionally-managed compute ecosystem.