Researchers have published CHAP (Collaborative Human-Agent Protocol), a new specification addressing the critical gap in how humans and AI agents collaborate in production environments. Published on arXiv on June 8, 2026, the protocol provides standardized infrastructure for accountable multi-human, multi-agent work affecting customers, claims, code, contracts, and clinical decisions.
Existing Standards Address Adjacent Problems But Miss Key Collaboration Layer
Two standards already exist for agent operations: MCP standardizes agent access to tools and data, while A2A standardizes agent-to-agent interoperability. However, neither defines the shared workspace where humans and agents perform accountable work together. Currently, the moment of human judgment—the most valuable signal in production systems—is recorded inconsistently across application code, chat threads, ticket comments, and tribal memory.
CHAP Defines Mandatory Core Elements and Optional Composable Profiles
The protocol consists of mandatory core elements including workspaces, participants, tasks, artefacts, and an append-only evidence log. Optional composable profiles cover review, modes, routing, deliberation, handoff, identity, signatures, and transparency-backed audit.
Key transformations include:
- Human edits of agent drafts become structured events with diff, rationale, and content hash
- Shift handoffs become portable envelopes rather than pinned messages
- Human approvals become non-repudiable signed decisions that can be replayed years later
Protocol Addresses Production Deployment Realities
As foundation models increasingly carry responsibility for operational work, production deployments involve multi-human, multi-agent collaborations across teams and time zones. CHAP provides the missing infrastructure layer as AI moves from response generation into operational roles that require planning across steps and coordinating with other agents.
The specification, reference implementation, conformance suite, and worked examples are available at https://github.com/BrightbeamAI/chap. The protocol aims to establish a weakly-specified technical surface for human-agent collaboration that has been missing from production AI deployments.
Key Takeaways
- CHAP protocol published on arXiv June 8, 2026, standardizes human-agent collaboration in production systems with accountable audit trails
- Existing standards MCP and A2A cover agent-to-tool and agent-to-agent communication but not human-agent collaboration workspaces
- Protocol includes mandatory core elements (workspaces, participants, tasks, artefacts, evidence log) and optional composable profiles for review, routing, and signatures
- Human actions like edits and approvals become structured, non-repudiable events with content hashes that can be replayed years later
- Reference implementation and conformance suite available at https://github.com/BrightbeamAI/chap