Regulation map
What the RBI asks of an agent.
RBI FREE-AI, the June 2026 draft model-risk guidance and fair practices on recovery, mapped to elan1 controls for banks and NBFCs. Free to read, no form.
- Last reviewed
- Access
- Read free · no form
In short
The short answer.
Does RBI's draft model-risk guidance cover AI agents?
Yes, as drafted. The draft published for consultation on 24 June 2026 treats any system whose outputs materially affect decisions as a model, AI included. An agent that triages alerts or proposes a limit change is in scope, and the bank or NBFC that deploys it stays accountable for what it does.
- Applies to
- Proposed to apply across regulated entities, from banks to NBFCs in every layer. It is a draft: the final text may change, and nothing in it applies until the RBI issues it.
- Key dates
- Aug 2025The RBI's FREE-AI committee report: 18% of Indian regulated entities keep AI audit logs.
- 24 Jun 2026Draft Guidance on Regulatory Principles for Model Risk Management published for consultation.
Clause by clause
Each duty, and the control that answers it.
| Clause, and what it asks | elan1 control | Mapped to | Claim IDs |
|---|---|---|---|
| RBI FREE-AI framework, Aug 2025The Framework for Responsible and Ethical Enablement of AI. Its committee found that 18% of Indian regulated entities keep AI audit logs. | Govern receipts: every agent action, allowed, approved or refused, leaves a signed, hash-chained receipt your auditor checks offline. |
| |
| Draft model-risk guidance: human oversightHuman-in-the-loop or on-the-loop arrangements, the ability to override or suspend a model, a kill switch, and periodic human review of AI-driven decisions. | An agent proposes a change; elan1 Govern allows it, asks a person you name, or refuses it, before anything reaches the core system. Its emergency stop pauses every agent at once and starts the incident report. |
| |
| Draft model-risk guidance: records that outlast the modelDocumentation kept through the model's life, with records of decommissioned models proposed to be kept for 10 years or more. | Each decision is recorded with the agent's identity, the rule and the approver, in a form a validator can check offline, and kept for as long as your policy requires. |
| |
| Draft model-risk guidance: explainability where it mattersCredit underwriting is treated as material. Where a model cannot be fully explained, compensating controls such as output checks, closer monitoring and limits on use. | elan1 Decide routes alerts and documents by calibrated confidence and treats credit and anti-money-laundering outcomes as recommend-only: it suggests, a person decides. |
| |
| RBI fair practices on recovery | elan1 Resolve with the collect1 pack, which checks contact hours, frequency and wording before a message goes. Every money-moving step passes elan1 Govern. |
| No claim ID |
Mappings are drafts for your counsel and auditors. They are not certification, legal advice or regulator approval.
What this map does not claim.
- None of this replaces your own model risk framework or independent validation; it is designed to give both something to check.
By industry
Where this rule shows up.
Banking, Financial Services and Insurance
RBI FREE-AI framework, Aug 2025
Govern receipts.
- elan1 Govern
Banking, Financial Services and Insurance
RBI draft model risk guidance, Jun 2026
Govern stop switch, named approvers, receipts.
- elan1 Govern
Banking, Financial Services and Insurance
RBI fair practices on recovery
Resolve with the collect1 pack.
- elan1 Resolve
Read more
The explainer, and the official text.
Regulation watch
What RBI's draft model-risk guidance asks of an AI agent
A kill switch, a person in the loop and records kept for 10 years. What it means for agents.
6 min read
Paper · June 2026 · 3 min read
A record a stranger can check: designing an audit trail for agents
Why an agent's record should be verifiable offline, without trusting the vendor.
elan1 research team
Read free · no form
Questions
RBI, plainly.
No. The draft Guidance on Regulatory Principles for Model Risk Management, 2026 was issued for public consultation on 24 June 2026. The final text may change, and nothing in it applies until the RBI issues it. It is proposed to apply across regulated entities, from banks to NBFCs, and covers AI and ML models, so this map reads it as a draft.
The draft asks for documentation kept through a model's life, and proposes that records of decommissioned models be kept for 10 years or more. Vendor contracts must give access to technical documentation and audit. For an agent, the record should hold the input, the confidence, the rule and the approver for each action.
Actions, not only outputs. The RBI draft asks for a kill switch and the ability to override or suspend a model. Stopping an agent means stopping what it is about to change in the core, at once, without leaving half-finished work behind. elan1 Govern's emergency stop pauses every agent at once and starts the incident report.
No. None of this replaces your own model risk framework or independent validation; it is designed to give both something to check. elan1 Govern leaves a signed, hash-chained receipt for every agent action, allowed, approved or refused, with the rule and the person who approved it, and your validators and auditors can verify those receipts offline without trusting elan1.
In India by default: AWS Mumbai, three availability zones, no US region. elan1 Run also runs in your data centre or on an appliance. Personal fields are masked before any model call.
No. Mappings are drafts for your counsel and auditors. They are not certification, legal advice or regulator approval. Each row names the clause or duty, what it asks, the elan1 control that answers it and, where one is assigned, the claim ID behind it, so your counsel and auditors can test each mapping against the receipts.
Other maps
The other rules we map.
- Allowed · Approved by Asha R.
- Held for CFO
- Refused · rule 4.2