RBI wants banks to have an AI ‘kill switch’: Here’s what it means for customers

Amid rising unease over how artificial intelligence is being used in banking, the Reserve Bank of India has proposed that banks and all other regulated entities must be able to instantly override, suspend or deactivate any AI model used in their operations, including through a kill switch arrangement. According to a report by The Economic Times, the requirement is part of a wide-ranging draft framework on Model Risk Management released for public consultation.

The draft framework calls for strong human oversight of all AI-driven decision-making and says no AI model should operate without the ability to be shut down immediately if it produces harmful or erroneous outputs. It also says banks must guard against automation bias, or the tendency of employees to rely too heavily on AI outputs without applying their own judgment, and requires customer-facing AI systems to tell users that they are interacting with AI and to offer them the option of switching to a human at any stage.

The framework introduces a risk-based tiering structure under which regulated entities will have to classify all models by risk level, from simple spreadsheet-based calculators to complex frontier AI systems, and apply proportionate oversight, validation and controls. The risk tier of every model will have to be reviewed at least once a year. High-risk models will need approval from the Risk Management Committee of the Board before deployment and cannot be cleared only by technology or risk teams.

DON'T MISS THIS | RBI governor calls rate-hike talk ‘premature’ as geopolitical uncertainty persists

For the first time, the RBI has placed AI and model governance directly at the board level. Every regulated entity will need a Board-approved Model Risk Management Framework covering all models, whether they are developed in-house, sourced from vendors or built through a combination of both. The board will be responsible for approving the entity's risk appetite for model risk, setting policies for model risk tiering, and ensuring that these remain forward-looking and are informed by stress testing and scenario analysis.

The draft guidelines also take a firm position on third-party models, including AI platforms and models procured from fintech and technology vendors. The RBI has said that a regulated entity will remain fully accountable for the outcomes of any model it uses, irrespective of whether it built the model itself or sourced it externally. The central bank has also flagged supply chain risk, arising from over-dependence on a limited number of AI model providers, as an issue that banks must actively manage.

For AI and machine learning models, the RBI has proposed requirements beyond conventional validation. Banks will have to set explainability thresholds so they can explain, in simple terms, why a model produced a particular output. The draft also links these requirements to fairness and bias in AI-led decision-making.

The push for kill switch arrangements comes alongside the RBI's separate proposal on digital payments, which suggests a single facility allowing customers to instantly disable all digital payment transactions from their account. In its discussion paper on safeguards in digital payments to curb frauds, the RBI said such a mechanism could help reduce online fraud, drawing on a model used in Singapore, though it also noted implementation and customer-use challenges.