Written Submission to the UN Global Dialogue on AI Governance

“Governance is written as policy, but AI is released as software.” How do we move from high-level ethical declarations to technical guardrails that engineering teams can actually deploy? In our official written submission for the UN Global Dialogue on Artificial Intelligence Governance, Sunil Khemka (Technical Convenor – AI & Cybersecurity at OpenBharat.net, Indian Society of Artificial Intelligence and Law) outlines practical, software-first solutions for global AI governance. 🔑 Key Deliverables Proposed To move past abstract principles, the submission advocates for three practical outputs: 🏗️ Governance ‘Built Into Software’ Reference Design: An open, industry-standard blueprint that wires safety checks, bias testing, and audit logging directly into software release pipelines. 🛑 Standardized Automatic Emergency Stop: A universal minimum standard for automated kill-switches to halt rogue autonomous agents operating at machine speed. 📚 Repository of Enforced Governance Measures: A searchable, global database focused on enforced national policies rather than non-binding strategy documents. 💡 Core Priorities & Emerging Risks Open-Source as True Capacity Building: Redefining capacity building away from basic workshops toward open models, open data, and shared compute infrastructure to bridge the global AI divide. Reimagining Oversight: Distinguishing between nominal “human-in-the-loop” setups and meaningful, automated pre-execution verification for multi-agent chains. Governing Autonomous Operators: Addressing accountability in multi-party service chains and standardizing cross-border audit logs for autonomous AI systems acting without human intervention.