Bias and Discrimination in the Agentic Web and How Project NANDA Can Support Mitigation

The traditional Internet of the Web was designed for static web infrastructure and human-mediated interactions. The agentic web transitions this to a trillion-scale, highly dynamic, decentralized environment where autonomous agents discover, authenticate, and collaborate at machine speed. Project NANDA's Internet of AI Agents (IoA) represents a fundamental shift in computational architecture, moving from DNS-centric discovery to decentralized registries, verifiable metadata, and rapid resolution. This paper positions the ethical risks of the agentic web---bias amplification, power asymmetries, and covert manipulation---as systemic issues that emerge from scale, autonomy, and protocol design. We outline how NANDA's tools, including AgentFacts, verifiable credentials, adaptive resolvers, reputation systems, and DAO-inspired governance, can mitigate risk while preserving decentralization. We also propose a threat-model-informed audit checklist for protocol-induced and orchestration bias, and report its systematic application to 100 publicly available MCP server repositories, revealing pervasive gaps (89--98\% Unclear'' orNo'') in fairness, security, and accountability primitives at the orchestration layer, while identifying open challenges in traceability, sustainability, and fairness enforcement.