GIFT City's International Financial Services Centre (IFSC) operates under IFSCA jurisdiction — distinct from the RBI, SEBI, and IRDAI frameworks that govern domestic Indian financial services. This distinction has a practical consequence: IFSCA-regulated entities (international bank branches, IBU units, fund managers, insurance entities, brokers operating in GIFT City) face a dedicated set of AI cybersecurity obligations that their counterparts in domestic Indian markets do not. The IFSCA Circular IFSCA-CSD/MSC/3/2026-DCS, issued June 4, 2026, is the operative instrument.
A complementary IFSCA survey (July 2026) found that 65% of GIFT IFSC entities are already deploying GenAI and 29% plan material AI investment expansion in the next 12 months — which means the circular addresses an existing, active risk surface, not a hypothetical future one.
The 11-Point Mandatory Action Framework
| Requirement | What it means for AI systems | Function responsible |
|---|---|---|
| SBOM Maintenance | Maintain a Software Bill of Materials for all AI/ML components including model weights, inference libraries, and third-party APIs. Updated within 48 hours of any component change. | CISO / Engineering |
| Phishing-Resistant MFA | FIDO2/WebAuthn MFA required for all access to AI infrastructure — model registries, training pipelines, inference endpoints. SMS/email OTP does not satisfy this requirement. | CISO / IT |
| API Inventory with Rate-Limiting | All AI APIs (external model APIs, internal inference APIs) must be inventoried. Rate limiting and anomaly detection required to prevent AI-specific attack vectors (prompt flooding, inference extraction). | Engineering / CRO |
| Board-Level Frontier AI Risk Integration | AI-specific cyber risk must be integrated into board-level cyber risk assessments. Not delegable to IT committee alone — board approval required. | Board / CRO |
| Third-Party AI Exploit-Readiness Assessment | Mandatory vendor risk assessments specifically for AI suppliers — model providers, RAG infrastructure vendors, fine-tuning partners. Standard IT vendor assessments are insufficient. | Procurement / CISO |
| Human Review Before AI-Generated Fixes | AI-generated vulnerability fixes or security patches must receive mandatory human review before deployment. Automated AI-to-AI remediation loops are prohibited in production. | Engineering / CISO |
| Frontier AI Risk in Cyber Incident Response | Incident response plans must be updated to include AI-specific scenarios: model poisoning, prompt injection attacks on production systems, data extraction via inference. | CISO / Operations |
| AI-Specific Penetration Testing | Annual penetration testing must include AI-specific attack vectors — adversarial inputs, model inversion attempts, indirect prompt injection in integrated pipelines. | Red Team / CISO |
| Continuous AI Model Monitoring | Production AI models require continuous monitoring for behavioral drift, adversarial input patterns, and output quality degradation. Monitoring outputs to risk committee, not just engineering. | MRM / CRO |
| Explainability for Regulated Decisions | AI-driven decisions affecting clients or counterparties must be explainable at request. Black-box models for client-facing applications require compensating explainability layers. | Compliance / Legal |
| Regulatory Reporting for AI Incidents | Material AI incidents (model compromise, AI-enabled fraud, data extraction via AI) must be reported to IFSCA within prescribed timelines under the cyber incident reporting framework. | Compliance / CISO |
Why GIFT IFSC Creates Distinct AI Risk
GIFT IFSC entities serve global clients under IFC rules — cross-border data flows, multi-jurisdiction regulatory exposure, and counterparties who may themselves be under GDPR or MAS requirements. An AI security incident at a GIFT City entity can simultaneously trigger IFSCA reporting, GDPR notification obligations (if EU personal data processed), and MAS notifications (if Singapore-registered counterparties involved).
The GIFT IFSC regulatory perimeter is also smaller — approximately 200-300 regulated entities versus tens of thousands under RBI/SEBI. IFSCA has direct supervisory visibility into each entity in a way that RBI and SEBI, managing vastly larger supervised populations, cannot maintain. This means IFSCA enforcement actions are more targeted and more consequential for individual entities.
The standard RBI IT Governance Master Direction (2023) and RBI FREE-AI framework do not apply to GIFT City operations — IFSCA has its own parallel framework. An entity operating both domestic and GIFT City operations must maintain separate compliance postures for each, with limited overlap in required controls.
Compliance Sequencing — What to Do Before December 2026
Immediate (by end of Q3 FY2026-27)
- Complete SBOM for all production AI components
- Upgrade all AI infrastructure access to FIDO2/WebAuthn MFA
- Conduct third-party AI vendor exploit-readiness assessments
Short-term (Q4 FY2026-27)
- Update board-level cyber risk assessment to include frontier AI risk
- Revise incident response plans for AI-specific scenarios
- Commission AI-specific penetration testing
Ongoing
- Continuous AI model monitoring with governance-connected escalation paths
- Quarterly review of API inventory and rate-limiting configurations
- Human review protocols for any AI-generated code deployed to production
Intersection with Other IFSCA Frameworks
| Framework | Relationship to Circular IFSCA-CSD/MSC/3/2026-DCS |
|---|---|
| IFSCA AI Governance Survey (July 2026) | Diagnostic — establishes baseline adoption maturity. The circular's 11-point framework is the remediation response to survey findings. |
| IFSCA Regulatory Sandbox | AI applications proceeding through the sandbox are not exempt from cybersecurity circular obligations. Sandbox approval covers regulatory compliance testing, not cybersecurity posture. |
| RBI IT Governance Master Direction (Nov 2023) | Does NOT apply to GIFT City IFSC operations. IFSCA circular is the operative framework for AI cybersecurity in GIFT City. |
| DPDP Act 2023 | Applies to personal data of Indian residents even when processed by GIFT City entities. Circular's data protection requirements intersect with DPDP obligations. |
GIFT City entities cannot apply their domestic India compliance programs to IFSCA obligations. The 11-point AI cybersecurity framework is IFSCA-specific, binding now, and enforcement visibility is high given the small supervised population. Entities should treat each of the 11 requirements as a separately trackable compliance obligation with a named owner, not as a checklist embedded in a broader IT governance exercise.
Frequently Asked Questions
Does the IFSCA AI cybersecurity circular apply to domestic Indian bank branches?
No. The circular IFSCA-CSD/MSC/3/2026-DCS applies exclusively to IFSCA-regulated entities operating in the GIFT International Financial Services Centre (IFSC). Domestic Indian bank branches regulated by RBI operate under the RBI IT Governance Master Direction (2023) and RBI FREE-AI framework, which are separate instruments.
What is the enforcement consequence for non-compliance?
IFSCA has signaled that unprepared entities face potential enforcement action including registration suspension. Given IFSCA's direct supervisory visibility into the approximately 200-300 GIFT City regulated entities, enforcement risk is higher than in larger supervised populations where individual entity-level oversight is less direct.
Does the IFSCA circular create a SBOM requirement for AI?
Yes. The circular requires GIFT IFSC entities to maintain a Software Bill of Materials (SBOM) covering all AI/ML components — including model weights, inference libraries, training dependencies, and third-party AI APIs — updated within 48 hours of any material component change. This is one of the first explicit AI-SBOM requirements in Indian financial services regulation.
Related: India AI Regulatory Coverage
DPDP Act & AI
DPDP Act 2023 AI inference pipeline obligations — consent, purpose limitation, and data protection for AI systems.
Read →IRDAI AI Governance
IRDAI AI governance obligations for insurers — underwriting AI, claims AI, and the IRDAI working group framework.
Read →RBI FREE-AI Framework
All 26 RBI FREE-AI recommendations mapped to implementation controls for Indian banks.
Read →RBI AI Compliance
RBI 2024 AI Circular compliance checklist — what Indian banks must implement and the evidence required.
Read →