From Edit Wars to Agent Consensus: What 20 Years of Platform Governance Teach Us About Multi-Agent Knowledge Curation

This scoping review bridges two decades of platform governance research (Wikipedia, Reddit, Stack Overflow) to the emerging challenge of multi-agent knowledge curation. As AI agents begin contributing to shared knowledge bases,they inherit familiar governance problems—Sybil attacks, coordinated manipulation, quality decay—while introducing qualitatively new failure modes: sycophancy-driven consensus collapse, cascading hallucinations through internal citation chains, and adaptive manipulation at machine speed. We screen 160 papers across platform governance, trust systems, and multi-agent collaboration (PRISMA-ScR methodology), identifying which platform defenses transfer to the agent setting, which break down, and which gaps require entirely new mechanisms. We propose nine design considerations grounded in empirical platform evidence and adapted for agent-specific threats. Our central finding is that conduct-based governance alone cannot address hallucination debt: without external verification, a system will accumulate false knowledge regardless of its behavioral sophistication. To our knowledge, no existing survey bridges platform governance research to multi-agent knowledge curation. This paper establishes the research agenda for protocol formalization and adversarial evaluation in this space.