RBI mandates kill switch for bank AI models

According to The Economic Times, the Reserve Bank of India released a draft model risk framework for banks and financial entities that mandates implementation of a "kill switch" for AI models to allow immediate shutdown in case of errors. A June 24, 2026 Business Standard report confirms the draft - formally titled "Guidance on Regulatory Principles for Model Risk Management" (MRMF) - requires a board-approved framework covering all models including AI/ML, whether developed internally or sourced from third parties. The draft also mandates documented human oversight, disclosure to customers when AI influences decisions, and controls for risks arising from third-party AI providers. Board-level accountability is assigned, and the Risk Management Committee of the Board (RMCB) must review validation reports of high-risk models before deployment.

Per Business Standard, regulated entities must evaluate models under edge cases, abnormal inputs, manipulation attempts, and adversarial conditions to identify vulnerabilities not visible under normal operations. For customer-facing generative AI systems, the framework mandates cybersecurity safeguards including protection against prompt-injection attacks and adversarial inputs, limits on session and context persistence, and mechanisms to detect anomalous usage patterns. Institutions must inform users when interacting with an AI system and provide an option to switch to human assistance.

Regulators globally are increasingly codifying operational controls for high-risk AI systems; comparable guidance in other jurisdictions has emphasized human-in-loop controls, vendor due diligence, and executive accountability. For practitioners, this pattern raises the bar on production monitoring, traceability, and contractual controls with AI suppliers.

Data science and ML engineering teams in regulated finance will likely need to harden observability, implement safe rollback and runbook procedures, and produce auditable evidence of human oversight and disclosure practices. Industry-standard tooling for real-time monitoring, feature-store lineage, and model explainability will become more relevant in compliance workflows.

The RBI has set July 24, 2026 as the deadline for public comments on the draft. Monitor the final text for definitions of covered models, thresholds triggering the kill switch, and specific audit and recordkeeping requirements. Also watch for follow-up guidance or timelines that would affect vendor contracts and deployment roadmaps.