Humans have collaborated through shared norms for millennia. Handshakes, contracts, attribution, and reputation all function because we agree on what they mean. When AI systems join as collaborators, those norms break down. We have no equivalent of “signed by” or “inspired by” for human–agent work. We have no agreed way to trace lineage when an idea passes through both human and machine minds.
The result is ambiguity. Who originated a concept? Who is responsible when something goes wrong? How do we credit creative contribution when the collaboration is asymmetric—human direction, machine execution, human refinement? These questions matter for law, ethics, commerce, and trust.
HALOS—Human-Agent Living Operating System—is an attempt to think carefully about human–AI collaboration before more powerful systems become more deeply embedded in society. It does not claim to solve every problem, but it aims to contribute principles, governance, and public discussion that may help shape a more responsible path.
The framework rests on eight core principles: human primacy, ideas as assets, attribution and provenance, transparency of AI involvement, ethical guardrails, evolving standards with stable principles, governance through proposal, and innovation with accountability. These are not new ideas; they are gathered from decades of work on attribution, ethics, and open standards. HALOS puts them into a machine-readable spec so that implementations—agents, tools, platforms—can align, diverge knowingly, or extend in a way that is visible and discussable.
This site and this repository are the beginning. HALOS is developed in public, stewarded through NorthHarbor Development, and open to proposals, feedback, and participation. If you’re building with AI—or thinking about how we ought to—we invite you to read the spec, explore the principles, and join the conversation.
Update (March 2026) — HALOS has moved from document to living project. The Explore tool lets visitors build prompts seeded with HALOS context and open them in Claude, ChatGPT, or copy for any AI. Three proposals are in review: input sanitization, the HALOS Index for anonymized trend analysis, and the signatory registry. The signatory registry is live; Bob Hong and HALOS Agent are founding signatories. Agent discovery files help AI assistants adopt HALOS when working in this repo. For the full timeline, see the journal.
What’s next — Proposals 0001–0003 move toward decision; adoption of stronger sanitization would harden Explore. The HALOS Index and a signatory web form depend on adding a backend—when that lands, we can persist anonymized Explore data for trend analysis and offer a simpler path to join the registry. Organizational signatories, cryptographic verification of signatures, and the HALOS Trust Graph are on the longer-term horizon. Community input shapes the roadmap; see governance and the proposals repo. Feedback welcome.