RBI Proposes Mandatory AI 'Kill Switch' for Banks
Amid growing concerns over artificial intelligence applications in the financial sector, the Reserve Bank of India (RBI) has introduced a draft framework requiring banks and other regulated entities to implement an 'AI kill switch.' This measure would enable the instant override, suspension, or deactivation of any AI model operating within their systems, particularly if it generates harmful or incorrect results.
According to the proposed Model Risk Management framework, which is now open for public consultation, no AI model should function without the explicit ability to be shut down immediately. This initiative underscores the central bank's commitment to mitigating risks associated with advanced AI deployment in critical financial operations.
Ensuring Human Oversight and Accountability
A cornerstone of the RBI's new guidelines is the insistence on robust human oversight for all AI-driven decision-making processes. The framework aims to combat 'automation bias,' where employees might over-rely on AI outputs without applying independent judgment. Banks will be required to:
- Inform customers when they are interacting with an AI system.
- Offer customers the option to switch to human support at any stage of an AI-led interaction.
Furthermore, the RBI has clarified that regulated entities will bear full accountability for the outcomes of any AI model they utilize, regardless of whether it was developed in-house or sourced from third-party vendors. This extends to managing supply chain risks associated with over-dependence on a limited number of AI model providers.
Risk-Based Tiering and Board-Level Governance
The draft framework introduces a comprehensive risk-based tiering structure for all models, ranging from simple spreadsheet tools to complex frontier AI systems. Each model's risk level must be reviewed annually. High-risk models will necessitate approval from the Board Risk Management Committee before deployment, moving beyond clearances from technology or risk teams alone.
For the first time, the RBI is placing AI and model governance directly at the board level. Every regulated entity must establish a Board-approved Model Risk Management Framework covering all models. The board will be responsible for defining the entity's risk appetite for model risk, setting policies for model risk tiering, and ensuring these are forward-looking and informed by stress testing and scenario analysis.
Explainability and Bias Mitigation
Beyond conventional validation, the RBI has proposed specific requirements for AI and machine learning models. Banks will need to set 'explainability thresholds,' enabling them to articulate, in simple terms, why a model produced a particular output. This requirement is directly linked to addressing issues of fairness and bias in AI-led decision-making, ensuring transparency and equitable treatment for customers.