Source-linked AI summary
ClawNet: Human-Symbiotic Agent Network for Cross-User Autonomous Cooperation
Zhiqin Yang, Zhenyuan Zhang, Xianzhang Jia, Jun Song, Wei Xue, Yonggang Zhang, Yike Guo
TL;DR
Existing agent systems automate individual tasks but lack infrastructure and governance for agents representing different users in collaboration. The paper proposes a human-symbiotic paradigm and instantiates it in ClawNet, whose layered identities and governance primitives support owner-bound, auditable cooperation.
Problem
Existing agent frameworks serve a single user and do not provide mechanisms for agents representing different owners to collaborate with identity, authorization, and accountability.
Method
The paper proposes permanently owner-bound agent systems with a privacy-preserving manager agent, context-specific identity agents, scoped authorization, and action-level accountability.
Results
ClawNet instantiates the paradigm as an identity-governed framework and validates governance-aware agents through representative cross-organizational collaboration scenarios.
Takeaways & Limitations
Cross-user autonomous cooperation is framed as digitizing human collaborative relationships through governance primitives that preserve identity isolation, authorization enforcement, and auditability.
Abstract
from arXiv · showhide
Current AI agent frameworks have made remarkable progress in automating individual tasks, yet all existing systems serve a single user. Human productivity rests on the social and organizational relationships through which people coordinate, negotiate, and delegate. When agents move beyond performing tasks for one person to representing that person in collaboration with others, the infrastructure for cross-user agent collaboration is entirely absent, let alone the governance mechanisms needed to secure it. We argue that the next frontier for AI agents lies not in stronger individual capability, but in the digitization of human collaborative relationships. To this end, we propose a human-symbiotic agent paradigm. Each user owns a permanently bound agent system that collaborates on the owner's behalf, forming a network whose nodes are humans rather than agents. This paradigm rests on three governance primitives. A layered identity architecture separates a Manager Agent from multiple context-specific Identity Agents; the Manager Agent holds global knowledge but is architecturally isolated from external communication. Scoped authorization enforces per-identity access control and escalates boundary violations to the owner. Action-level accountability logs every operation against its owner's identity and authorization, ensuring full auditability. We instantiate this paradigm in ClawNet, an identity-governed agent collaboration framework that enforces identity binding and authorization verification through a central orchestrator, enabling multiple users to collaborate securely through their respective agents.
1. Introduction
Existing AI agents automate tasks for individual users but lack infrastructure for agents to represent distinct owners in collaboration. ClawNet addresses this gap with a human-symbiotic paradigm, layered identities, governance primitives, and an implemented framework.
- Research gap: Existing single- and multi-agent systems serve one principal and provide no mechanism for agents representing different users to interact.Their agents share unified goals, while human productivity depends on coordination, negotiation, and delegation across organizational relationships.
- Research gap: Interoperability protocols provide agent communication but do not bind agents to owners, enforce authorization scopes, or create auditable responsibility chains.
- Proposed paradigm: ClawNet proposes permanently owner-bound agent systems with multiple context-specific identity agents for different collaborative relationships.The paradigm focuses on digitizing human collaborative relationships rather than only strengthening individual agent capability.
- Proposed paradigm: A layered architecture isolates the globally knowledgeable manager agent from external communication while exposing context-specific identity agents to collaborators.Each identity agent represents only the information relevant to its designated collaborative context.
- Implementation and validation: ClawNet instantiates the paradigm as an identity-governed framework with cloud-edge execution, scoped file authorization, and operational auditing.Its contributions include identity binding, scoped authorization, action-level accountability, and representative cross-organizational collaboration scenarios.
2. System Overview
ClawNet models collaboration as a human-node network whose agent-mediated edges are governed by identity, authorization, and accountability. Its cloud-edge architecture combines persistent orchestration with controlled local file execution.
- Human-symbiotic network: ClawNet defines a collaboration network whose nodes are human users and whose edges are governed relationships mediated by their agents.This differs from multi-agent topologies in which agents themselves are the primary network actors.
- Layered identity: Each user owns a permanently bound agent system containing a manager agent and multiple identity agents.The manager governs identity-agent creation, configuration, and retirement, holds complete owner knowledge, and remains isolated from external communication.
- Layered identity: Each identity agent represents a collaborative context through scoped authorization, context-relevant knowledge, and an authorized collaborator set.Only users in that authorized set can discover and interact with the identity.
- Cross-user collaboration: Cross-user collaboration requires explicit contact and identity assignment, bilateral authorization, bounded delegation, and enforcement against crossing lower-level authorization boundaries.The collaboration lifecycle includes contact establishment, multi-turn dialogue, and manager consultation.
- Governance: Three governance constraints require identity binding, scoped authorization, and append-only action logging for every operation.Authorization violations are intercepted and escalated to the owner, while logs record the operation, owner, identity, result, and time.
- Cloud-edge architecture: The cloud-edge design keeps reasoning and identity state persistent for collaboration while enabling OS-level execution on the owner’s device.Cloud components include gateway containers and a central orchestration engine; the edge client provides interaction, auditing, and local file operations.
3. Identity Governance and Agent Collaboration
ClawNet binds agents to owners through persistent cognitive coupling and governs cross-user collaboration with layered identities, scoped knowledge, and human-held authority.
- 3.1. Identity Binding: From Task Memory to Cognitive Coupling: ClawNet builds owner understanding through factual, pattern, and value memory accumulated progressively across interactions.These layers cover owner facts, behavioral habits, communication styles, decision boundaries, and value dispositions.
- 3.1. Identity Binding: From Task Memory to Cognitive Coupling: Persisting these memory layers across sessions transforms a generic utility into a personalized owner proxy.The paper calls this persistent alignment cognitive coupling.
- 3.2. Layered Identity Architecture: A layered architecture separates a globally informed manager agent from outward-facing identity agents carrying context-relevant knowledge.The manager agent is isolated from external communication, while identity agents represent specific collaborative roles.
- 3.2. Layered Identity Architecture: Each identity agent is a contextual projection defined by its role, authorization boundary, knowledge subset, and authorized collaborators.Identity agents can be instantiated, retired, or reconfigured as the owner’s social roles change.
- 3.3. Cross-User Collaboration: Cross-user collaboration preserves human authority over initiation, identity selection, information-sharing boundaries, and final decisions.Agents handle information exchange, deliberation, and progress synchronization after the owner sets these boundaries.
4. Related Work
Related work advances individual agents, multi-agent coordination, governance, and interoperability, but does not provide ClawNet’s complete cross-user identity-governance framework.
- Single-agent and multi-agent frameworks: Single-agent systems provide OS-level control for one user, while multi-agent frameworks coordinate agents that share a principal and goal.These systems do not represent distinct human interests in collaboration.
- Agent safety and governance: Agent-safety research identifies governance, identity, delegation, and accountability needs but does not supply the same implemented framework.The cited works address governance practices, authenticated delegation, or accountability gaps at different levels.
- Agent interoperability protocols: Interoperability protocols standardize agent discovery, messaging, or tool integration without establishing owner binding, authorization scopes, or responsibility chains.They address communication rather than the full governance of cross-user representation.
5. Conclusion
ClawNet instantiates human-symbiotic cross-user cooperation by binding agents to owners, separating private and outward-facing identity tiers, and enforcing governance throughout interactions.
- 5. Conclusion: ClawNet permanently binds each agent to a specific human owner through progressive cognitive coupling.This binding supports agents acting as representatives rather than generic utilities.
- 5. Conclusion: The framework partitions identity into privacy-preserving and outward-facing tiers for cross-user cooperation.The conclusion presents this partition as part of the human-symbiotic paradigm.
- 5. Conclusion: ClawNet enforces identity binding, scoped authorization, and action-level accountability throughout interactions.The conclusion frames these primitives as infrastructure for digitizing human collaborative relationships.
- 5. Conclusion: The paper argues that autonomous agents will need governance infrastructure before interacting with other people, organizations, and institutions.This conclusion extends the framework’s motivation beyond individual agent capability.