A researcher has published two companion papers introducing OpenAgenet (OAN), an open infrastructure project designed to establish trusted identity and verification standards for AI agents operating across multi-operator networks — addressing a gap that becomes critical as AI agents move beyond isolated applications into interconnected, autonomous ecosystems.
As artificial intelligence agents grow more capable of acting independently and communicating with one another, a foundational question has emerged: how can one agent safely trust another it has never encountered before? Jinliang Xu, the author behind the OpenAgenet project, published two technical papers on 4 June 2026 outlining a proposed solution through an open, protocol-neutral trust layer.
The OpenAgenet project — abbreviated as OAN — does not seek to replace existing agent interaction standards. Instead, it positions itself as an underlying trust layer that operates beneath protocols such as MCP (Model Context Protocol), A2A (Agent-to-Agent), and ANP (Agent Network Protocol). The system is designed to answer a specific set of questions before any substantive agent interaction begins: Is this agent who it claims to be? Is it authorised to participate in this network? Is its governance state current and valid?
According to the papers, OAN achieves this through several interlocking components. A Root-governed identity admission system controls which agents are permitted to join the network. A Registrar-assisted onboarding process manages how new agents are introduced and verified. The system also incorporates Root-verified package publication, ensuring that software components agents rely upon can be traced back to authorised sources. Discovery within the network is authorisation-aware, meaning agents can only be found by those permitted to interact with them, and invocations are cryptographically signed to provide verifiable evidence of intent.
One notable design element is a blockchain-backed authorisation bulletin, which the papers describe as a mechanism for maintaining a tamper-evident, distributed record of agent authorisation states. This allows participants in the network to verify the current governance standing of any agent without relying on a single centralised authority.
The architecture is explicitly designed to support heterogeneous frameworks — meaning agents built on different underlying platforms or using different interaction protocols could still participate in the same trusted network. The papers describe OAN as defining "how Agent identities become admissible, discoverable, verifiable, and safe to approach" rather than governing the substance of what agents do once connected.
Xu's papers describe the project as currently at prototype stage, with a performance profile and roadmap included in the first publication. The work represents an academic and engineering response to what the author characterises as an increasingly visible gap: the absence of a shared, open trust infrastructure as AI agent deployments become more distributed and commercially significant.
The research arrives amid growing industry attention to agent interoperability standards, with major technology companies and open-source communities developing competing frameworks for how AI agents should communicate and coordinate.
Analysis
Why This Matters
- As AI agents are increasingly deployed in enterprise and consumer settings to perform autonomous tasks, the lack of a shared identity and trust standard creates real security and operational risks — a rogue or compromised agent in a networked system could cause significant harm.
- OAN represents one of the early serious attempts to define open, governance-aware infrastructure for agent networks, potentially influencing how industry standards evolve over the next several years.
- If widely adopted, such a trust layer could become foundational infrastructure for multi-agent AI systems, similar to how TLS/SSL became standard for web security.
Background
The rapid proliferation of large language model-based AI agents has outpaced the development of standards for how those agents should interact safely. Early agent deployments were largely self-contained — a single agent performing a defined task within one organisation's infrastructure. However, as commercial interest has grown, so has the vision of agents from different providers and operators working together: booking travel, executing financial transactions, coordinating logistics, or conducting research across organisational boundaries.
This shift has exposed a structural gap. Existing agent interaction protocols such as Anthropic's MCP, Google's A2A framework, and the community-developed ANP define how agents communicate, but not who is permitted to communicate or how trust between previously unknown agents should be established. Traditional approaches to identity — such as OAuth tokens or API keys — were not designed with autonomous, multi-hop agent chains in mind.
Blockchain-based authorisation mechanisms have been explored in previous decentralised identity projects (such as W3C's Decentralised Identifiers standard), and OAN appears to draw on that lineage while applying it specifically to the AI agent context.
Key Perspectives
Proponents of open standards: An open, protocol-neutral trust layer like OAN could prevent fragmentation, where each major platform creates its own incompatible trust system. Open infrastructure tends to produce more scrutiny and, ultimately, more robust security than proprietary alternatives.
Enterprise and commercial operators: Organisations deploying AI agents at scale have a strong interest in verifiable governance states and authorisation trails, particularly for regulated industries. A standardised trust layer could simplify compliance obligations and reduce liability exposure from agent misbehaviour.
Critics and sceptics: The project is currently at prototype stage and originates from a single researcher, raising questions about whether it can achieve the breadth of adoption necessary to become a genuine standard. Blockchain-backed components may introduce complexity and latency concerns. Competing proposals from well-resourced organisations may crowd out independent academic proposals regardless of technical merit.
What to Watch
- Whether major AI platform providers (Anthropic, Google DeepMind, OpenAI, Microsoft) reference or engage with OAN's architecture in their own agent interoperability roadmaps.
- The release of a production-ready implementation or reference deployment, which would signal whether the project moves beyond academic proposal toward practical adoption.
- Emerging regulatory requirements around AI agent accountability in the EU AI Act's implementation guidance and equivalent US frameworks, which could create demand for exactly the kind of verifiable governance trail OAN proposes.