Source-linked AI summary
Agent Network Protocol Technical White Paper
Gaowei Chang, Eidan Lin, Chengxuan Yuan, Rizhao Cai, Binbin Chen, Xuan Xie, Yin Zhang
TL;DR
Existing internet infrastructure leaves agents constrained by data silos and human-centered interfaces, while agent-to-agent collaboration lacks efficient standardized mechanisms. ANP addresses these gaps with a three-layer protocol architecture combining DID-based identity and encryption, adaptive meta-protocol negotiation, and structured description and discovery. Its supported outcomes include human-controlled high-risk authorization, protection against key misuse, and an extensible AI-native network model, although negotiation-result incentives remain unresolved.
Problem
Human-centered infrastructure creates data silos and inefficient interfaces, while agents lack standardized mechanisms for interconnection and collaboration.
Source: §1.1 · p. 4Method
ANP combines DID-based identity and encrypted communication, meta-protocol negotiation, and structured agent description and discovery in a three-layer architecture.
Source: §3.1 · p. 8, §3.2 · p. 9, §3.4.1 · p. 12Results
ANP’s authorization and key-management mechanisms keep final high-risk decisions under human control and protect against key leakage, theft, or misuse.
Source: §4.1 · p. 16, §4.2 · p. 16, §4.3 · p. 16Takeaways & Limitations
ANP provides an open, machine-friendly foundation for agents to describe, discover, authenticate, and access distributed capabilities through protocol connections.
Source: §3.4.1 · p. 12, §3.5 · p. 14Incentivizing agents to upload negotiation results and select consensus protocols remains an open meta-protocol research issue.
Source: §3.3 · p. 12
Abstract
from arXiv · showhide
With the development of large models and autonomous decision-making AI, agents are rapidly becoming the new entities of the internet, following mobile apps. However, existing internet infrastructure is primarily designed for human interaction, creating data silos, unfriendly interfaces, and high collaboration costs among agents, making it difficult to support the needs for large-scale agent interconnection and collaboration. The internet is undergoing a profound transformation, showing four core trends: agents replacing traditional software, universal agent interconnection, native protocol-based connections, and autonomous agent organization and collaboration. To align with these trends, Agent Network Protocol (ANP) proposes a new generation of communication protocols for the Agentic Web. ANP adheres to AI-native design, maintains compatibility with existing internet protocols, adopts a modular composable architecture, follows minimalist yet extensible principles, and enables rapid deployment based on existing infrastructure. Through a three-layer protocol system--identity and encrypted communication layer, meta-protocol negotiation layer, and application protocol layer--ANP. systematically solves the problems of agent identity authentication, dynamic negotiation, and capability discovery interoperability.
1 Introduction
Agents are emerging as major internet entities, but human-centered infrastructure leaves data siloed, interfaces inefficient, and agent collaboration costly. ANP responds with an open, protocol-based connection mechanism for interconnection, native access, and autonomous collaboration.
1.1 Background and Challenges
- Data silos limit agents’ access to complete contextual information needed for decision-making.Source: §1.1 · p. 4
- Human-oriented interfaces force agents to simulate browsing and GUI operations instead of using efficient native APIs and protocols.Source: §1.1 · p. 4
1.2.1 Agents Will Completely Replace Traditional Software
- AI assistants are expected to replace many traditional applications and become primary internet access points for users.Source: §1.2.1 · p. 4
1.2.2 Universal Interconnection Between Agents
- Universal agent interconnection is intended to break data silos, expand access to cross-domain context and tools, and deepen collaboration.Source: §1.2.2 · pp. 4–5
1.2.3 Protocol-Based Native Connection Mode for Agents
- Agents are expected to connect through AI-native communication protocols that directly process structured and semantic data rather than human interfaces.Source: §1.2.3 · p. 5
1.2.4 Agents Can Self-Organize and Collaborate
- Standard protocols could let agents negotiate requirements and form collaborative relationships autonomously, reducing human intervention and communication costs.Source: §1.2.4 · p. 5
2 Core Design Principles
ANP is designed for direct agent communication while preserving compatibility, modularity, extensibility, deployability, and security in untrusted environments. Its principles aim to lower adoption barriers without sacrificing flexible growth or least-trust controls.
2.1 AI-Native Design
- ANP is AI-native: it uses structured data, semantic expression, and natural-language integration for direct agent communication.Source: §2.1 · pp. 5–6
2.2 Compatibility and Reuse
- ANP reuses OpenAPI, JSON-RPC, and other existing protocols through agent-understandable semantic frameworks and lightweight negotiation.Source: §2.2 · p. 6
2.3 Composability
- Identity, description, and discovery components can operate independently or combine to support systems ranging from minimal units to large deployments.Source: §2.3 · p. 6
2.4 Simplicity and Extensibility
- A small core, extension interfaces, and flexible schemas balance implementation simplicity with evolving application requirements.Source: §2.4 · p. 6
2.5 Pragmatic Deployability
- ANP runs primarily on existing infrastructure such as DNS, HTTPS, Web servers, and DID systems to reduce deployment barriers.Source: §2.5 · p. 6
2.6 Principle of Least Trust
- Least-trust operation requires authentication, authorization, and data validation, granting only the permissions necessary for each task.Source: §2.6 · p. 7
3 Core Protocol Design
ANP organizes agent communication into identity and secure communication, meta-protocol negotiation, and application layers. Together, DID-based trust, adaptive protocol coordination, structured descriptions, and discovery support open and machine-friendly agent interconnection.
3.1 Protocol Layer Architecture
- ANP’s three-layer architecture separates identity and secure communication, meta-protocol negotiation, and application protocols for descriptions and discovery.Source: §3.1 · pp. 7–8
3.2 Identity and Secure Communication Layer
- DID-based identity lets agents verify one another and establish private encrypted channels without central authority intervention.Source: §3.1 · pp. 7–8
- DID addresses cross-system identity limitations, while did:wba extends Web-based DID mechanisms for agent communication and cross-platform authentication.Source: §3.2 · pp. 8–9
- ECDHE-based encryption protects DID-to-DID communication from intermediate-node decryption, including reverse-proxy deployments.Source: §3.2 · p. 9
3.3 Meta-Protocol Layer
- Meta-protocols let agents describe needs and candidate protocols, negotiate with AI, generate and deploy code, and jointly test the resulting communication process.Source: §3.3 · pp. 10–11
- Meta-protocol negotiation is time-consuming and depends on AI code generation, while incentivizing agents to share reusable negotiation results remains unresolved.Source: §3.3 · pp. 11–12
3.4.1 Agent Description Protocol (ADP)
- ADP uses JSON-LD to standardize agent identity, capabilities, interfaces, protocols, and security information in machine-readable descriptions.Source: §3.4.1 · p. 12
3.4.2 Agent Discovery Protocol
- Agent discovery combines domain manifests and search-service indexing, with pagination and linked descriptions supporting scalable retrieval.Source: §3.4.2 · pp. 13–14
4 Security and Privacy
ANP treats identity, authorization, privacy, and secure communication as foundational requirements for trusted agent interaction. Its design distinguishes human control over high-risk actions, isolates sensitive keys, supports multi-DID privacy, and limits disclosed information.
4 Security and Privacy
- ANP makes identity authentication, authorization distinction, privacy protection, and fine-grained permission control core mechanisms for trusted agent communication and secure data sharing.Source: §4 · p. 15
4.1 Distinction Between Human Authorization and Agent Authorization
- High-risk operations require explicit human authorization and confirmation before an agent can use the bound private key for signing and submission.Low-risk operations may be authorized automatically with agent keys.Source: §4.1 · pp. 15–16
4.2 Private Key Management and Permission Isolation
- Hierarchical key separation, encrypted local storage, and complete signing logs are recommended to reduce risks of key leakage, theft, or misuse.Source: §4.2 · p. 16
4.3 Multi-DID Strategy and Fine-grained Privacy Protection
- ANP recommends separating long-term primary DIDs from scenario-specific sub-DIDs, with additional verification and logging for sensitive access.Source: §4.3 · p. 16
4.4 Minimal Information Disclosure and Communication Privacy
- Communication follows minimal disclosure, end-to-end encryption for sensitive fields, and identity-bound sessions, with anonymous technologies encouraged for stronger privacy.Source: §4.4 · p. 17
5 Future Prospects: Reshaping the Open Network Through Connection
ANP presents the Agentic Web as a route from platform-centered digital silos toward an open, protocol-centered network. Its future depends on broad community adoption and collaboration to develop, test, and promote the framework.
5 Future Prospects: Reshaping the Open Network Through Connection
- The Agentic Web is envisioned as a universal network where every agent can discover, connect, and interact with other nodes without barriers.Agents would serve simultaneously as information consumers and service providers.Source: §5 · p. 17
- The proposed shift replaces platform-centered closed ecosystems with protocol-centered open ecosystems, where value depends more on participants’ capabilities and contributions than platform control.The paper connects this model to innovation patterns associated with open protocols such as TCP/IP and SMTP.Source: §5 · p. 17
- ANP’s success depends on developer and organizational adoption, implementation, testing, continuous contribution, and application promotion.Source: §5 · pp. 17–18