Does RBI MRMF apply to AI used in PMJDY overdraft scoring at public sector banks?

Yes. RBI MRMF June 2026 applies to all AI models used by regulated entities in material financial decisions, regardless of the customer segment. A BSBDA overdraft scoring model used to determine the overdraft limit for a Jan Dhan account holder is a material AI model — it influences a credit decision that affects the customer's access to emergency credit. MRMF Section 3 (validation), Section 4 (explainability), and Section 6 (ongoing performance monitoring) all apply. Public sector banks that have deployed overdraft scoring AI for their PMJDY portfolio without formal model validation, without explainability for declined overdraft applications, or without production monitoring are not MRMF-compliant. The financial inclusion mission does not exempt these models from MRMF — if anything, the vulnerability of the beneficiary population and the government's policy intention that overdraft access be genuinely available create a stronger governance obligation, not a lighter one.

What are the INDIC-BIAS testing obligations for AI models serving Jan Dhan account holders?

INDIC-BIAS testing for PMJDY AI models requires assessing whether the model produces systematically different outcomes for the population segments that Jan Dhan account holders predominantly represent: first-time banking customers without credit bureau history, low-income wage earners and daily workers, rural and semi-urban residents, women (who constitute over 55% of the PMJDY account base), and beneficiaries from Scheduled Caste and Scheduled Tribe communities. AI models trained on the historical behavior of established banking customers may have systematic gaps in their performance for bureau-thin populations — not because of intentional discrimination but because the training data does not adequately represent first-time banking customers. INDIC-BIAS testing should measure: false positive rate (incorrectly denied access to overdraft or services) and false negative rate (incorrectly approved access that creates customer distress) across each of these segments, and compare them to the overall population's rates. Elevated false positive rates for any of these segments relative to the overall rate indicate a bias that requires model remediation under FREE-AI Sutra 4 (Fairness and Equity).

What AI governance obligations apply to DBT fraud detection at public sector banks?

Direct Benefit Transfer (DBT) fraud detection AI at PSU banks operates at the intersection of two governance frameworks: RBI MRMF (which applies because the bank is an RBI-regulated entity) and MeitY/DFS joint guidelines on DBT fraud analytics (which apply because the bank is operating as a disbursing agency for government scheme funds). The MeitY/DFS guidelines require PSU banks to implement fraud flagging criteria aligned with the DBT Mission's defined typologies: duplicate beneficiary detection, ghost beneficiary filtering, and redirected payment detection. AI models used for these functions must be configured to flag the specific patterns the DBT Mission guidelines define — a model that detects general transaction fraud but has not been calibrated for DBT-specific fraud typologies does not satisfy the MeitY/DFS guidelines. Additionally, false positives in DBT fraud detection have severe consequences for beneficiaries: a Jan Dhan account flagged as fraudulent by an AI model may have its DBT disbursements held, denying a vulnerable person access to PMJJBY insurance proceeds, scholarship transfers, or LPG subsidy — outcomes that create legal liability under consumer protection law and political accountability for the public sector bank.

What governs AI used in automated processing of PMJJBY and PMSBY insurance claims?

Pradhan Mantri Jeevan Jyoti Bima Yojana (PMJJBY) and Pradhan Mantri Suraksha Bima Yojana (PMSBY) claims processing AI at public sector banks is subject to two regulatory layers. The IRDAI layer: IRDAI Insurance Digital Distribution Regulations 2024 and IRDAI's AI Working Group recommendations apply to insurance claim processing AI, requiring disclosure when AI is involved in claims assessment and human review for disputed claims. The RBI layer: for PSU banks acting as insurance intermediaries or claim-processing agents, RBI's customer protection requirements and MRMF apply to the AI systems they operate. A PSU bank that uses AI to assess PMJJBY death claims — validating nominee identity, checking for duplicate claims, assessing document authenticity — must have that AI governed under MRMF. The specific risk: AI models that incorrectly reject legitimate PMJJBY claims for Jan Dhan account holders' families, denying ₹2 lakh death benefit to a low-income family, creates both regulatory liability and social harm at scale.

AI Governance · PMJDY · Financial Inclusion · PSU Banks · DBT · India BFSI

PMJDY Financial Inclusion AI Governance: Jan Dhan Overdraft Scoring, DBT Fraud Detection, and PMJJBY Claim AI at PSU Banks

530+ million Jan Dhan accounts. Public sector banks deploy AI for BSBDA overdraft scoring, DBT fraud detection, dormancy prediction, and PMJJBY automated claim processing. RBI MRMF applies to every one of these models. False positives for Jan Dhan AI have consequences that go beyond model risk — they deny vulnerable beneficiaries access to the financial inclusion safety net the government built.

Zytra Research Team · October 2026 · 13 min read

📖 Part of the India BFSI AI Governance Complete Guide 2026 — RBI, SEBI, IRDAI, NPCI & DPDP in one place.

Pradhan Mantri Jan Dhan Yojana has 530+ million accounts, disbursed over ₹2.3 trillion in Direct Benefit Transfers, and brought over 300 million adults into the formal banking system for the first time. SBI alone operates over 250 million Jan Dhan accounts. AI governs how these accounts are scored, whether their overdraft facilities are activated, whether their DBT disbursements are processed or held on fraud flags, and — increasingly — whether their linked insurance claims are approved or rejected.

This is not a niche application of AI governance. It is the largest financial inclusion program in human history, run through the AI systems of India's public sector banks, operating under RBI MRMF's governance requirements. A governance failure in PMJDY-related AI does not affect a statistically sampled customer population — it affects specific Jan Dhan account holders whose access to overdraft credit, government benefits, or insurance proceeds depends on a model working correctly for their profile.

No global AI governance vendor has written about PMJDY AI governance. This post maps the specific AI deployments, the regulatory frameworks that govern them, and the governance obligations that PSU bank compliance and risk teams must address.

The AI Stack in PMJDY Operations

PMJDY's operational AI deployments span five distinct functions at PSU banks. Each has different governance obligations, different false-positive consequences, and different regulatory oversight.

1. BSBDA Overdraft Scoring

The PMJDY scheme provides eligible account holders access to an overdraft facility of up to ₹10,000 (subject to satisfactory account operation). The eligibility criteria — six months of satisfactory account operation, no history of delinquency on other credit obligations — are rule-based, but PSU banks have deployed AI models to supplement these criteria with behavioral scoring: account activity patterns, salary/pension credit regularity, DBT receipt history, mobile banking usage.

Under MRMF, these overdraft scoring models are material — they influence a credit decision that affects a Jan Dhan account holder's access to emergency credit at a moment of financial stress (the overdraft facility exists precisely for emergencies). MRMF Section 3 requires validation; Section 4 requires instance-level explainability (a declined overdraft application must be explainable to the customer); Section 6 requires ongoing performance monitoring.

The INDIC-BIAS obligation is particularly important here: overdraft scoring models trained on historical customer data may have systematic gaps for the population PMJDY targets — first-time banking customers, rural workers with irregular income patterns, women whose account activity reflects household management patterns rather than individual economic activity. A model that performs well on the overall PMJDY population but systematically denies overdraft to women, rural residents, or tribal community members has failed FREE-AI Sutra 4 even if its aggregate accuracy metrics look acceptable.

2. DBT Fraud Detection AI

Direct Benefit Transfer fraud detection sits at the intersection of banking AI governance and government program AI governance. PSU banks act as disbursing agencies for dozens of government schemes — PMJJBY, PMSBY, PM-KISAN, PM-KISAN Samman Nidhi, scholarship programs, LPG subsidies — and are required by the DBT Mission and MeitY to implement fraud detection for these disbursements.

The MeitY/DFS joint guidelines on DBT fraud analytics (2023) define three core fraud typologies that PSU banks must configure their fraud detection systems to address:

Duplicate beneficiary fraud: multiple accounts receiving the same benefit entitlement, often using different identity documents for the same individual or family
Ghost beneficiary fraud: benefits disbursed to accounts linked to deceased individuals or to accounts with no real corresponding beneficiary
Redirected payment fraud: legitimate beneficiary accounts that have been compromised and are used to redirect benefits to fraudulent accounts

AI models used for DBT fraud detection must be calibrated to detect these specific typologies, not just general transaction anomalies. A model that flags unusual transaction amounts or velocity may miss ghost beneficiary fraud entirely — ghosts receive regular, normal-amount benefit disbursements that look unremarkable in isolation.

The false-positive consequence in DBT fraud detection is unusually severe. A Jan Dhan account incorrectly flagged as fraudulent by an AI model may have its disbursements held pending review. For a family dependent on PMJDY-linked scholarship payments or PM-KISAN agricultural support, a hold of even a few weeks during a financial stress period can cause material harm. PSU banks must govern their DBT fraud AI false-positive rates with the same rigor as credit AI, and must have an expedited review process for held disbursements that does not require weeks of manual processing.

3. Dormancy Prediction AI

Dormant PMJDY accounts — accounts that have not transacted for 24 months — are classified as inoperative under RBI guidelines and require additional procedures to reactivate. PSU banks use AI to predict which accounts are approaching dormancy thresholds, enabling proactive outreach to account holders before their accounts become inoperative.

While this is a less acute governance risk than overdraft scoring or DBT fraud detection, dormancy prediction AI still requires MRMF compliance: the model must be documented, validated, and monitored. Additionally, dormancy prediction models may be used for cross-sell and account retention decisions — proactively offering the account holder additional services to increase engagement. If the dormancy prediction model's output is used to target marketing or service offers, the marketing decision becomes an AI-influenced decision requiring MRMF governance and DPDP consent for the use of personal data in marketing targeting.

4. PMJJBY and PMSBY Automated Claim Processing

Pradhan Mantri Jeevan Jyoti Bima Yojana (PMJJBY) provides a ₹2 lakh life insurance cover; Pradhan Mantri Suraksha Bima Yojana (PMSBY) provides ₹2 lakh accidental death and disability cover. Both are bundled with Jan Dhan accounts. PSU banks process the claims through their insurance intermediary or corporate agent operations.

AI in PMJJBY/PMSBY claim processing performs several functions: nominee identity verification (biometric matching against Aadhaar for death claims), document authenticity assessment (death certificates, medical reports for PMSBY disability claims), duplicate claim detection, and claim amount calculation verification. Each of these is a material AI function under MRMF when the AI output influences whether a claim is approved or referred for manual review.

The regulatory governance for PMJJBY/PMSBY claim AI is layered:

IRDAI Insurance Digital Distribution Regulations 2024: disclosure that AI is involved in claims processing; human review for disputed claims; claim rejection explanation
MRMF: the PSU bank's AI systems for claim processing are material models requiring validation and monitoring
Consumer Protection Act 2019: incorrect AI-driven claim rejection is a deficiency in service — the PSU bank bears civil liability for AI-generated rejections of legitimate claims

5. Simplified KYC Biometric AI

PMJDY accounts use simplified KYC provisions — biometric verification through Aadhaar authentication is the primary KYC mechanism for Jan Dhan account holders without full documentation. AI is used in the biometric authentication process (liveness detection, face matching against Aadhaar photograph) and in ongoing KYC compliance monitoring (periodic re-verification, biometric update tracking).

Biometric AI governance for PMJDY accounts must address CERT-In's guidance on deepfake threats to biometric verification — the vKYC bypass attacks that CERT-In documented in the Digital Threat Report 2025-26 are particularly relevant for PMJDY's Aadhaar-based authentication, which relies on the same video and biometric verification infrastructure that is being targeted by deepfake attacks.

Multi-Regulator Governance Framework for PMJDY AI

PMJDY AI governance sits at the intersection of multiple regulatory authorities — a complexity that is unique to this topic and that requires PSU banks to manage a multi-regulator compliance structure:

AI Function	Primary Regulator	Secondary Regulator	Key Obligation
BSBDA Overdraft Scoring	RBI (MRMF)	Ministry of Finance (DFS)	MRMF validation + INDIC-BIAS + adverse action explanation
DBT Fraud Detection	MeitY/DFS (DBT Mission)	RBI (MRMF)	DBT typology coverage + false-positive governance + held-disbursement review SLA
Dormancy Prediction	RBI (MRMF, KYC Master Direction)	DPDP (if used for marketing)	Model documentation + DPDP consent for marketing use
PMJJBY/PMSBY Claim AI	IRDAI (Insurance Distribution)	RBI (MRMF for PSU bank systems)	Claim rejection explanation + human review for disputed claims + MRMF validation
Simplified KYC Biometric AI	RBI (KYC Master Direction)	CERT-In (deepfake defense)	Liveness detection calibration + deepfake testing + UIDAI Aadhaar authentication compliance

PMJDY AI Governance Checklist

☐ BSBDA overdraft AI model: in MRMF model inventory; model card complete; independent validation conducted
☐ Overdraft AI explainability: instance-level explanation generated for declined applications; plain-language adverse action notice available on request
☐ INDIC-BIAS testing: overdraft model false-positive rate segmented by gender, geography (rural/semi-urban/urban), bureau-thin vs bureau-positive, SC/ST status — compared to overall rates; elevated rates investigated
☐ DBT fraud detection: model calibrated for MeitY/DFS typologies (duplicate beneficiary, ghost beneficiary, redirected payment) — not generic transaction fraud
☐ DBT false-positive SLA: held disbursement review process with defined SLA; expedited review for DBT holds during distress periods (post-natural disaster, agricultural seasons)
☐ DBT false-positive segmentation: false-positive rate monitored by scheme type, geographic region, and beneficiary demographic — elevated rates escalated
☐ PMJJBY/PMSBY claim AI: IRDAI disclosure requirement implemented; human review pathway for disputed rejections; claim rejection explanation mechanism
☐ PMJJBY/PMSBY claim AI MRMF: model documented, validated, monitored; false positive rate (incorrect rejection) tracked as primary metric
☐ Dormancy prediction AI: model card complete; if used for marketing targeting, DPDP consent documented for marketing data use
☐ Simplified KYC biometric AI: liveness detection calibrated for AI-generated video (CERT-In deepfake advisory); Aadhaar authentication UIDAI API compliance; INDIC-BIAS for biometric performance across demographic groups
☐ Multi-regulator compliance mapping: all PMJDY AI functions mapped to applicable regulatory frameworks; accountable owner identified for each framework
☐ FREE-AI Sutra 4 governance: documented policy that PMJDY AI systems are subject to fairness assessment; INDIC-BIAS results reviewed at governance committee level

PMJDY AI governance is not a niche concern for a small fintech — it is a large-scale operational governance challenge for every PSU bank in India. SBI, PNB, Bank of Baroda, and Canara Bank collectively operate over 350 million Jan Dhan accounts. Each AI model decision that affects those accounts — a declined overdraft, a held DBT disbursement, a rejected PMJJBY claim — affects a real person whose financial resilience depends on the system working correctly. MRMF's governance requirements create the framework for ensuring it does.