Collaborative Human-Agent Protocol (CHAP)

A new protocol standard called the Collaborative Human-Agent Protocol (CHAP) has been proposed to define the shared workspace where humans and AI agents perform accountable work together, addressing a gap left by existing standards, according to a paper submitted in 2026 [1]. The paper, hosted on the open-access repository arXiv, argues that foundation models are transitioning from simple response generation into operational roles where they plan, use tools, and coordinate with other agents [1]. The authors note that production deployments are no longer limited to one human supervising one model, but now involve multi-human, multi-agent collaborations that cross teams, time zones, and trust boundaries [1]. Despite this shift, the technical surface for human-agent collaboration remains weakly specified [1]. When a human edits an agent's draft before it is finalized, that moment of judgment is described as the most valuable signal in the system, yet it is often recorded only in application code, chat threads, or ticket comments [1]. Two existing protocol standards address adjacent concerns: MCP standardizes agent access to tools and data, and A2A standardizes agent-to-agent interoperability, but neither defines a shared workspace for joint accountable work [1]. CHAP is designed to fill this void. Under the protocol, a human override of an agent's output is captured as a structured event carrying a diff, a rationale, and a content hash, rather than disappearing into a chat thread [1]. A handoff between shifts becomes a portable envelope instead of a pinned message, and a human's approval of an agent's draft becomes a non-repudiable signed decision that can be replayed years later [1]. The protocol achieves this through a small Core consisting of workspaces, participants, tasks, artefacts, and an append-only evidence log, along with composable profiles that add features such as review, routing, deliberation, handoff, identity, signatures, and transparency-backed audit as deployments require them [1]. The specification, a reference implementation, a conformance suite, and worked examples are available on GitHub [1]. arXiv, which began on August 14, 1991, and passed the two-million-article milestone by the end of 2021, serves as the primary distribution platform for this proposal [10].