Quick answer
RBI's circular of 6 April 2018 requires payment system providers to store the entire payment data only in India. Data for the foreign leg of a transaction may be processed abroad, but RBI's June 2019 FAQs say it must be brought back and deleted abroad within one business day or 24 hours, whichever is earlier. AI workloads on payment data must respect this; DPDP section 16(2) keeps it in force.
- RBI's 6 April 2018 circular requires the entire payment system data to be stored only in India.
- Per RBI's June 2019 FAQs, data processed abroad must return to India and be deleted abroad within one business day or 24 hours, whichever is earlier.
- Training or running AI models abroad on payment data generally conflicts with the rule unless it fits RBI's permitted processing conditions.
- DPDP section 16(2) preserves stricter sectoral rules, so payment data localisation survives DPDP's permissive transfer regime.
- Tokenised or in-country inference designs let payments companies use advanced AI without moving payment data abroad.
Who this is for: Payment aggregators and gateways, Card networks and payment system operators, Banks running payment systems, Fintechs with fraud and risk ML, Cloud and AI vendors serving payments companies
What does RBI's payment data localisation rule require?
On 6 April 2018, RBI issued a circular on storage of payment system data. It requires all system providers to ensure that the entire data relating to payment systems operated by them is stored in a system only in India. The obligation extends to data held by their service providers and intermediaries on their behalf.
RBI's FAQs of June 2019 clarified the edges:
- "Payment data" is read broadly: end-to-end transaction details and information collected, carried or processed as part of the payment instruction.
- For cross-border transactions, data relating to the foreign leg may also be stored abroad.
- Processing may happen outside India, but the data must then be stored only in India, and data abroad must be deleted within one business day or 24 hours of payment processing, whichever is earlier.
The rule predates the DPDP Act and is not displaced by it. Section 16(2) of the DPDP Act says that where any other law provides a higher degree of protection or restriction on cross-border transfer, that law applies.
Read the circular and FAQs themselves on rbi.org.in for definitions and any later clarifications before finalising architecture — and check the latest circular, as RBI may refine its position.
Which AI workloads does it affect?
Payments companies use ML and generative AI heavily. Each use case has to be checked against the storage-in-India rule.
| AI use case | Payment data involved? | Localisation impact |
|---|---|---|
| Real-time fraud scoring | Yes — transaction details, device, merchant | Model serving and feature stores in India |
| Model training for fraud or risk | Yes — historical transactions | Training data stays in India; training jobs in Indian regions |
| Chargeback and dispute copilots | Yes — transaction records | LLM must run in India or receive only non-payment data |
| Merchant analytics dashboards | Often yes | Aggregates may be less constrained; raw data stays in India |
| Customer support chatbots | Sometimes (transaction lookups) | Retrieval layer in India; minimise what goes to the model |
| Marketing models | Possibly | Check whether inputs are payment data |
| Code assistants for engineers | Usually no | Ensure no production payment data in prompts |
The question to ask: does the data in this prompt, feature or training set fall within RBI's description of payment data? If yes, storage must be in India and any foreign processing must meet the FAQ conditions. Most AI vendors' default architectures — logs, caches, fine-tuning stores and evaluation traces kept abroad for days or indefinitely — do not meet a 24-hour deletion window.
Architecture options for AI on payment data
Option 1: in-country everything
Run training, feature stores, model serving, vector databases and observability in Indian cloud regions or on-premises. Open-weight models make this practical for many tasks. This is the cleanest compliance position.
Option 2: tokenise and abstract before any foreign call
Where a foreign model API adds value (for example, summarising a merchant's written complaint), send only non-payment content and replace transaction identifiers with tokens resolved in India. Keep the token vault and the mapping in India. Rule 6(1)(a) of the DPDP Rules recognises virtual tokens mapped to personal data as a security measure.
Option 3: foreign processing under the FAQ conditions
If processing abroad is unavoidable, the vendor must store nothing beyond the processing window: no persistent logs, no caching, no training retention, deletion within one business day or 24 hours, whichever is earlier. Get contractual and technical evidence. Few general-purpose AI APIs offer this by default; zero-retention settings need verification.
Logging
Keep LLM gateway logs, audit trails and traces in India. That also satisfies CERT-In's 180-day logs-in-India direction and supports DPDP's one-year retention from 13 May 2027 (Rules 6(1)(e) and 8(3)). See CERT-In rules for AI systems.
Payment data localisation and the DPDP Act
Payment data about individuals is personal data under the DPDP Act. So payments companies carry two overlapping sets of duties.
- Transfer. DPDP section 16 is permissive (negative list), and Rule 15 adds conditions only on foreign State access. RBI's rule is stricter, and section 16(2) makes it prevail.
- Security. Rule 6 requires encryption, masking or tokenisation, access controls, logging and monitoring, backups and processor contract clauses — for payment data stored in India as much as anywhere.
- Breach. A breach of payment data involving individuals triggers Rule 7: intimation to each affected person and to the Board, with a detailed report within 72 hours. RBI and CERT-In reporting run in parallel.
- Retention. Rule 8(3) requires at least one year; RBI and other financial laws may require longer, and section 8(7) lets the longer legal retention govern.
- Processors. Section 8(1) keeps the payments company responsible for its vendors' processing; Rule 6(1)(f) requires contract clauses.
For the full sector picture, see DPDP for banks, NBFCs and fintechs, and for the broader hosting map, data localisation for AI workloads.
Vendor checks for AI tools used by payments companies
Ask each AI vendor these questions before payment data goes anywhere near it.
- Where is data stored and processed? Region list, including sub-processors, logging and support systems.
- What is retained, and for how long? Prompts, outputs, embeddings, logs, abuse-monitoring copies, backups.
- Can retention be set to zero or deletion within 24 hours? And is that contractual and verifiable?
- Is data used for training? It must not be, by default.
- Who can access it? Vendor staff locations and access controls.
- Incident notification. Fast enough to meet CERT-In's 6-hour window and DPDP's 72-hour Board report.
- Audit rights. Including RBI's right to inspect outsourced arrangements where applicable.
- Exit and deletion. Certified erasure on termination.
Common mistakes
- Enabling a vendor's default logging that stores prompts abroad for weeks.
- Sending raw transaction records to a foreign LLM for "just a summary".
- Keeping a foreign disaster recovery copy of payment data.
- Treating fraud model features as non-payment data when they are derived from transactions.
- Assuming DPDP's permissive transfer rule overrides RBI's circular.
What to do now
Payment data localisation has applied since 2018. What changes by 13 May 2027 is that DPDP security, breach, retention and rights obligations layer on top. A practical sequence:
- Inventory AI use cases that touch payment data, including copilots and support bots.
- Classify inputs — payment data, other personal data, non-personal data.
- Pick an architecture per use case from the three options above.
- Lock down vendors with region, retention, training and audit terms.
- Move logs and traces into India and set retention to at least one year.
- Rehearse an incident that triggers RBI, CERT-In and DPDP reporting together.
- Align with FREE-AI. RBI's August 2025 FREE-AI report recommends an AI inventory, risk classification and AI incident reporting — the same artefacts serve both.
Check RBI's site periodically for clarifications on payment data storage, as interpretations of edge cases (for example, aggregated analytics or derived data) can evolve.
Further reading on suhasbhairav.com
Before 13 May 2027
Compliance checklist
Identify payment data in AI flows
Tag every prompt, feature, embedding and training set that contains payment data.
RBI circular, 6 Apr 2018
Host payment-data AI in India
Run training, serving, vector stores and traces for payment data in Indian regions.
RBI circular, 6 Apr 2018
Tokenise before foreign calls
Send only tokens or non-payment content to foreign model APIs.
DPDP Rule 6(1)(a)
Enforce deletion abroad
Where foreign processing is used, verify deletion within one business day or 24 hours.
RBI FAQs, June 2019
Keep AI logs in India
Store gateway logs and audit trails in India for at least one year.
CERT-In Directions; DPDP Rule 8(3)
Update vendor contracts
Add region, retention, no-training, incident notification and audit clauses.
DPDP Rule 6(1)(f)
Plan joint incident reporting
Prepare one workflow covering RBI, CERT-In and DPDP Rule 7 reports.
DPDP Rule 7
Key dates
RBI circular on storage of payment system data
DPDP security, breach and retention obligations apply on top of RBI rules
Answers
Frequently asked questions
Can payment data be processed outside India?
Can payment data be processed outside India?
Yes, in limited form. RBI's June 2019 FAQs allow processing abroad, but the data must be stored only in India and any copy abroad deleted within one business day or 24 hours of processing, whichever is earlier. Foreign-leg data for cross-border transactions may also be stored abroad.
Can I train a fraud model on payment data in a foreign cloud?
Can I train a fraud model on payment data in a foreign cloud?
Generally no, because training requires storing data abroad well beyond the 24-hour window. Run training in an Indian region, or train abroad only on data that is not payment data under RBI's definition. Check the latest circulars and FAQs before deciding.
Does the DPDP Act override RBI's localisation rule?
Does the DPDP Act override RBI's localisation rule?
No. DPDP section 16(2) says that where another law provides a higher degree of protection or restriction on cross-border transfer, that law applies. RBI's payment data storage rule therefore continues alongside the DPDP Act's permissive negative-list approach.
Can a payments company use ChatGPT-style tools?
Can a payments company use ChatGPT-style tools?
Yes, for tasks that do not involve payment data, or where payment data is tokenised and nothing that RBI treats as payment data leaves India. For tasks that need payment data, use models hosted in India. Verify vendor retention and training settings before use.
What counts as payment data for AI purposes?
What counts as payment data for AI purposes?
RBI's FAQs describe payment data broadly as end-to-end transaction details and information collected, carried or processed as part of a payment message or instruction. Features derived from transactions, such as spending patterns, should be treated cautiously and assessed against that description.
Who is responsible if an AI vendor stores payment data abroad?
Who is responsible if an AI vendor stores payment data abroad?
The payment system provider. RBI's circular covers data stored by service providers on its behalf, and DPDP section 8(1) keeps the Data Fiduciary responsible for its processors. Contracts, configuration and audits need to prevent foreign retention.
Legal references and sources
- RBI circular on storage of payment system data — 6 April 2018; FAQs June 2019 (official text)
- Digital Personal Data Protection Act, 2023 — Section 16(2) (official text)
- Digital Personal Data Protection Rules, 2025 — Rule 6 and Rule 15 (official text)
This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.
About the author

Suhas Bhairav
Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.
Related guides
- India AI & sector rulesData Localisation in India: Where Can You Host AI Workloads?Where you can legally host AI workloads that touch Indian personal data: DPDP's negative list, Rule 13(4) for SDFs, RBI payment data, CERT-In logs and sectoral rules, in one decision table.
- Sector guidesDPDP for Banks, NBFCs and FintechsHow banks, NBFCs, payment companies and fintechs layer DPDP obligations on top of RBI rules: consent, retention, breach, cross-border, AI credit models and Consent Managers.
- India AI & sector rulesRBI FREE-AI Framework: What Banks & Fintechs Must DoWhat the RBI FREE-AI committee report asks of banks, NBFCs and fintechs, what is and is not binding yet, and how it maps onto DPDP obligations from May 2027.
- DPDP Rules 2025DPDP Cross-Border Data Transfer Under Rule 15India allows personal data to leave the country by default. Section 16 adds a negative list, Rule 15 adds Government orders on access by foreign States, and Rule 13(4) can localise data for SDFs.
- India AI & sector rulesCERT-In Rules for AI Systems: Incident Reporting & LogsHow CERT-In's April 2022 directions on 6-hour incident reporting, 180-day logs in India and clock sync apply to LLM gateways, model APIs and AI agents, and how they meet DPDP.
- AI & personal dataEncryption, Masking and Tokenisation for AI Pipelines Under DPDPRule 6(1)(a) names encryption, obfuscation, masking and virtual tokens as minimum safeguards. Here is how to apply each one to prompts, embeddings, training sets and model gateways.