RBI proposes ‘kill switch’ for AI systems used by banks

The Reserve Bank of India has proposed a comprehensive AI governance framework requiring banks to maintain emergency shutdown mechanisms, strengthen human oversight, classify AI models by risk, and enhance accountability for AI-driven decisions.

The Reserve Bank of India (RBI) has proposed a new regulatory framework aimed at strengthening oversight of artificial intelligence systems used by banks and financial institutions, including a requirement that all AI models be capable of being immediately suspended or deactivated when necessary.

Released for public consultation on Wednesday (June 24), the draft guidelines seek to establish a structured approach to managing risks associated with AI and machine learning technologies that are increasingly being deployed across the financial sector. The central bank has invited comments on the proposals until July 24.

Under the draft framework, regulated entities would be required to maintain mechanisms that allow them to override, halt or shut down AI systems if they generate incorrect, harmful or unintended outcomes. The proposal effectively calls for institutions to have an emergency control arrangement, often referred to as a “kill switch,” to ensure AI models do not continue operating without supervision during critical situations.

Human oversight and risk-based controls

The RBI has emphasized that AI-driven processes should not function without meaningful human supervision. Even when automated systems are used to support business operations or decision-making, financial institutions would remain responsible for ensuring that human oversight is maintained.

The proposed rules cover a broad spectrum of models, ranging from basic analytical tools and automated calculators to advanced artificial intelligence platforms. Banks and other regulated entities would be held accountable for the performance and outcomes of these models, irrespective of whether they were developed internally or sourced from external vendors.

To strengthen governance, the central bank has suggested a risk-tiering framework that would require institutions to classify AI models according to their potential impact. Models assessed as carrying higher levels of risk would be subject to stricter controls, validation processes and approval requirements before deployment.

The draft also states that if a model's risk level exceeds an institution’s tolerance threshold, corrective action must be taken promptly. Such measures could include restricting usage, enhancing controls, modifying the system or discontinuing its operation altogether. Risk classifications would need to be reviewed annually, while high-risk models would require approval from the board-level Risk Management Committee.

Board accountability and customer protection

In a significant governance shift, the RBI has proposed that every regulated entity establish a board-approved Model Risk Management Framework covering all AI and analytical models in use. The regulator noted that weaknesses in model governance could expose institutions to operational, financial, compliance and reputational risks.

The draft guidelines also address emerging concerns such as dependence on a limited number of AI providers, cybersecurity vulnerabilities and excessive reliance on automated recommendations. Banks would be expected to actively manage these risks and implement additional safeguards for customer-facing generative AI applications.

To improve transparency, institutions would need to inform customers when they are interacting with AI systems and provide an option to switch to a human representative whenever required.

The proposed framework reflects the RBI’s growing focus on ensuring that rapid AI adoption in the financial sector is accompanied by robust governance, accountability and consumer protection measures.