RBI Proposes AI Risk Framework for Banks: Kill Switch and Board Rules Explained
The Reserve Bank of India (RBI) has proposed a new AI risk framework for banks. The RBI is India’s central bank. It makes the rules that all banks and lenders must follow. On June 24, 2026, it released a draft rulebook called the “Guidance on Regulatory Principles for Model Risk Management, 2026.” In simple words, it tells banks how to safely use software models, including AI tools, when they make decisions about your money.
AI means artificial intelligence. It is computer software that learns from data and can make choices on its own. Banks now use AI to approve loans, spot fraud, and score credit. The RBI wants to make sure these tools are fair, safe, and under human control. The draft is open for public comments until July 24, 2026.
What is a “model” and why does the RBI care?
A “model” is any system that takes in data and gives an answer. For example, a model can look at your income and decide if you get a loan. Older rules only covered credit models. The new draft is much wider. It covers every model a bank uses in any part of its business.
This includes models the bank builds itself. It also includes models bought from outside companies. And it includes all AI and machine-learning tools. Machine learning (ML) is a type of AI that gets better as it sees more data. The rule applies even if the bank does not think of the tool as a “model.” This wide net is the biggest change in the draft.
The “kill switch” for AI
The most talked-about rule is the “kill switch.” A kill switch is an emergency stop button. If an AI model starts acting in a wrong or risky way, a human can switch it off at once. This keeps a person in charge, not the machine.
The RBI wants this human control built into every AI system. The draft lists seven AI risks that banks must watch. These include bias (unfair results), hallucinations (the AI making up false answers), and data problems. The kill switch is the safety net for when these risks appear.
The board must take charge
The draft makes top bosses responsible. Every bank must have a Model Risk Management Framework, or MRMF. This is a written plan approved by the bank’s board of directors. The board is the group of senior people who run the bank. The plan must cover a model’s full life, from build to shut-down.
High-risk models need extra sign-off. The Risk Management Committee of the Board must check the test reports of these models. It must then approve their use. Banks must also rank each model by risk. A risky model cannot be marked as “low risk” just because it looks simple. The RBI calls this an “anti-dilution rule.”
Outside vendors are no escape
Many banks buy AI tools from outside firms. The RBI says this does not remove the bank’s duty. The bank stays fully responsible for what the tool does. Even if the vendor says the tool is safe, the bank must test it on its own. This is called independent validation. Contracts with vendors must also allow audits and a clean exit.
Key facts
| Item | Detail |
|---|---|
| Draft name | Guidance on Regulatory Principles for Model Risk Management, 2026 |
| Released by | Reserve Bank of India (RBI) |
| Release date | June 24, 2026 |
| Public comment deadline | July 24, 2026 |
| Scope | All models across all business processes, including third-party and AI/ML models |
| Key safety rule | “Kill switch” to suspend or shut down AI models |
| Board duty | Board-approved Model Risk Management Framework (MRMF) |
| Records | Decommissioned models must be kept for at least 10 years |
FAQ
What is a kill switch for AI?
It is an emergency stop button. A human can switch off an AI model at once if it starts giving wrong or risky results. This keeps people in control of the machine.
Are these RBI rules final?
No. This is a draft. The RBI has asked banks and the public to send feedback by July 24, 2026. The final rules may change after that.
Does this build on an earlier RBI report?
Yes. It follows the RBI’s August 2025 FREE-AI report. FREE-AI stands for the Framework for Responsible and Ethical Enablement of Artificial Intelligence. That report set the ground for these rules.
Why it matters (especially for India / founders)
India is one of the fastest users of digital banking in the world. Banks here run on apps, UPI, and instant loans. So AI safety is a big deal for everyday people. These rules aim to stop unfair loan rejections and hidden errors in AI.
For founders, the message is clear. If you build AI tools for banks, you must prove they are safe and easy to audit. Banks will now demand strong testing and a clear exit plan. Fintech startups should design for this from day one. This links to wider RBI moves, such as its recent decision on Tata Sons and NBFC rules, which shows the RBI is tightening its grip across finance. The same care also matters for tech firms like Infosys, which is expanding its AI-led services.
The takeaway
The RBI wants AI in banking to be safe, fair, and under human control. The kill switch, board oversight, and vendor checks are the core ideas. Banks have until July 24, 2026, to respond. Once final, these rules will shape how AI runs inside every Indian bank.
Sources
- Financial Express — RBI proposes AI risk framework for banks
- CorpLawUpdates — RBI Draft Guidance on Model Risk Management 2026
- The420 — New RBI Guidelines Mandate Kill Switches for Banking AI
