What are the AI governance obligations in the RBI AI-ACT&RS Advisory 2026 for Indian banks and payment system operators?

The RBI Advisory on AI-Accelerated Cyber Threats and Related Safeguards (AI-ACT&RS) — issued April 27, 2026 to all commercial banks (including Small Finance Banks and Payments Banks) and June 1, 2026 to all authorised non-bank Payment System Operators — is the first RBI supervisory advisory to address AI and Generative AI governance as a distinct regulatory subject in India's banking system. Five core AI governance obligations: (1) Board-approved AI risk gap assessment (time-bound): every regulated entity must conduct and submit to the RBI a board-approved AI risk gap assessment against AI-ACT&RS requirements. The board-approval requirement is the critical governance dimension — it elevates AI risk from an IT department matter to a board-level accountability obligation. The June 30, 2026 deadline created an immediate compliance timeline; (2) Safe and governed use of AI and Generative AI: the advisory explicitly requires institutions to ensure that all AI and Generative AI (GenAI) tools deployed within the institution operate under documented governance frameworks covering model selection, deployment controls, output validation, and incident response. This is India's first RBI-issued document to specifically name Generative AI as a regulatory governance subject — prior RBI AI frameworks (FREE-AI, Agentic AI Guidance, MRMF 2026) addressed AI broadly; this advisory names GenAI explicitly; (3) AI-driven threat preparedness plan: institutions must develop a time-bound action plan for improving detection, response, and resilience against AI-enabled cyber attacks — including attacks from frontier AI models that identify previously undetected software vulnerabilities faster than human security teams can remediate them; (4) AI-driven social engineering controls: supplementary controls specifically targeting AI-driven social engineering — deepfake audio and video impersonation, AI-voice cloning for authorisation fraud, synthetic identity creation. Existing cybersecurity frameworks (the 2024 Master Circular on Cyber Security) are acknowledged as insufficient for these AI-specific attack vectors; the advisory requires supplementary controls beyond the existing framework; (5) CIS (Cyber Information Security) gap assessment for AI threats: institutions must formally assess gaps in their existing Cyber Information Sharing protections as these relate to AI-accelerated threats — a new threat category not covered by the 2024 CIS framework.

Why is the RBI AI-ACT&RS Advisory's specific naming of Generative AI significant for India's banking AI governance framework?

The explicit naming of Generative AI (GenAI) in the RBI AI-ACT&RS Advisory marks a regulatory milestone in India's banking AI governance framework. Three dimensions of significance: (1) First GenAI-specific banking regulation in India: RBI's prior AI frameworks — the FREE-AI Framework (principles-based AI governance), the MRMF 2026 (Model Risk Management Framework), and the Agentic AI Guidance 2025 — use the term 'AI' broadly, covering all algorithmic decision-making systems. None specifically named Generative AI or Large Language Models as a distinct regulatory category. The AI-ACT&RS Advisory's explicit 'Generative AI' reference creates a specific regulatory accountability obligation for GenAI deployment that prior frameworks did not carry — an institution that deploys GenAI under a general AI governance framework but without GenAI-specific documentation may be non-compliant with AI-ACT&RS; (2) Bidirectional AI risk framing: the advisory introduces a bidirectional AI risk framing absent from prior RBI frameworks. Prior RBI AI frameworks focused on internal AI risk — the risk that the bank's own AI makes bad decisions. AI-ACT&RS introduces external AI risk — the risk that adversarial AI is used against the bank. Generative AI enables both: a bank can deploy GenAI internally for customer service or document processing, and a threat actor can use GenAI to generate synthetic fraud documents or deepfake bank executive communications. Both dimensions are now within RBI's supervisory scope; (3) Interaction with India's broader AI policy landscape: the AI-ACT&RS Advisory is issued in the context of India's developing national AI governance architecture — including MeitY's AI governance framework discussions, the DPDP Act 2023's automated decision-making provisions, and SEBI's parallel AI/ML Responsible Usage Framework for capital markets. The RBI advisory's explicit GenAI framing aligns with India's recognition that foundation models (GPT-4 class, Gemini class) require governance attention distinct from traditional ML models — large-scale capability, opaque reasoning, and susceptibility to prompt injection attacks create risks that statistical model risk management does not address.

What AI-driven cyber attack vectors does the RBI AI-ACT&RS Advisory specifically require Indian banks to address?

The RBI AI-ACT&RS Advisory identifies AI-enabled cyber attack vectors that require supplementary controls beyond India's existing cybersecurity framework (the 2024 Master Circular on Cyber Security). Five specific AI-enabled threat categories: (1) Frontier AI vulnerability discovery: advanced AI models can identify software vulnerabilities at a speed and scale beyond human security team capacity — analyzing millions of lines of banking application code to find exploitable bugs faster than remediation patches can be deployed. Banks must assess their patch management and vulnerability scanning capabilities against AI-accelerated discovery timelines; (2) AI-driven deepfake impersonation: AI-generated video and audio deepfakes can impersonate bank executives, RBI officials, and senior management to authorize fraudulent transactions, override security controls, or manipulate staff. Supplementary verification controls for high-value authorizations and communications claiming to come from senior officials are required; (3) AI-voice cloning for telephonic fraud: AI voice cloning tools that replicate an executive's voice from minutes of publicly available audio create a new vector for social engineering attacks — fraudulent telephone authorizations for wire transfers, fund movements, or security override requests. Authentication controls must address AI-synthesized voice; (4) Synthetic identity fraud: AI-generated composite identities combining real data elements (Aadhaar number from one source, face image from another, income documentation from a third) can defeat KYC controls designed for document fraud rather than AI-assisted composite identity creation; (5) AI-assisted phishing at scale: LLM-generated phishing emails that are contextually accurate, grammatically correct, and personalized to the specific recipient — using publicly available information about the target — are significantly more effective than template phishing. Banks must update email security and phishing awareness training to address AI-generated content that bypasses traditional phishing detection.

How does the RBI AI-ACT&RS Advisory relate to and supplement India's existing RBI AI governance framework for banks?

The RBI AI-ACT&RS Advisory sits alongside several prior RBI AI governance instruments, addressing distinct dimensions that earlier frameworks did not cover: (1) Distinct from RBI FREE-AI Framework: the FREE-AI Framework (Fairness, Reliability, Explainability, Ethics and Accountability for AI) is a principles-based document covering the responsible deployment of AI within bank operations — the bank's own AI. AI-ACT&RS addresses both internal AI governance (GenAI deployment governance) and external AI threats (adversarial AI). The two frameworks are complementary and both binding; (2) Distinct from RBI Agentic AI Guidance 2025: the Agentic AI Guidance addresses the specific risks of autonomous AI agents executing multi-step banking workflows — authorization scope, tool use controls, human oversight checkpoints. AI-ACT&RS addresses frontier AI as a threat vector (AI used to attack the bank) and GenAI as a deployment category requiring explicit governance — different from the agentic workflow risk the 2025 guidance covers; (3) Extends the 2024 Master Circular on Cyber Security: the 2024 Master Circular establishes the comprehensive cybersecurity framework for banks (already covered in Zytra's corpus). AI-ACT&RS explicitly states that its requirements supplement — not replace — the 2024 Master Circular's obligations. Banks must comply with both simultaneously; AI-ACT&RS adds new obligations in the AI-specific cyber risk domain; (4) Interaction with RBI MRMF 2026: the Model Risk Management Framework 2026 covers AI model risk in bank operations — validation, monitoring, change management for models deployed in credit, fraud, and operations. GenAI models deployed by banks fall within MRMF scope as well as AI-ACT&RS scope. A bank deploying a GenAI customer service system must comply with MRMF 2026 (model risk management) and AI-ACT&RS (GenAI governance and threat preparedness) simultaneously.

AI Governance · RBI · GenAI · Cyber Threats · Board Assessment · Banks · PSOs · India BFSI

RBI Advisory on AI-Accelerated Cyber Threats and Related Safeguards (AI-ACT&RS) 2026: Board-Approved AI Risk Gap Assessment, India's First Generative AI Banking Regulation, AI-Driven Deepfake Safeguards and Time-Bound Threat Preparedness Plan for All Banks and Payment System Operators

RBI Advisory on AI-Accelerated Cyber Threats and Related Safeguards (AI-ACT&RS) — issued April 27, 2026 to all commercial banks (including Small Finance Banks and Payments Banks) and June 1, 2026 to all authorised non-bank Payment System Operators — is India's first RBI document to specifically name Generative AI as a banking regulatory governance subject. Board-approved AI risk gap assessment required by June 30, 2026. Safe and governed use of AI and GenAI tools mandated under documented frameworks. Supplementary controls required for AI-driven deepfake impersonation, AI-voice cloning, synthetic identity fraud, and frontier AI vulnerability discovery. Time-bound AI-driven threat preparedness action plan required. CIS gap assessment for AI-accelerated threats mandated beyond existing 2024 Master Circular obligations. Applies to all commercial banks and authorised non-bank PSOs. Zero global competitor coverage.

Zytra Research Team · April 2026 · 12 min read

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

Every prior RBI AI governance framework — FREE-AI, the Model Risk Management Framework, the Agentic AI Guidance — focused on the bank's own AI. The implicit assumption was that AI is something banks deploy and must govern responsibly. AI-ACT&RS changes that framing. AI is now also something used against banks — and the sophistication of those attacks, powered by frontier AI models, has exceeded what existing cybersecurity frameworks were designed to handle.

The board-approval requirement for the gap assessment is the regulatory statement that AI risk cannot remain in the IT department. A board that approves an AI risk gap assessment has taken on governance accountability for AI cyber preparedness — and the next time an AI-enabled fraud or cyber incident occurs, the board's own assessment of preparedness will be part of the regulatory record.

The GenAI Milestone

India's banking regulators have been careful to use 'AI' as a broad category rather than naming specific model architectures. AI-ACT&RS breaks that pattern by explicitly naming Generative AI. This matters operationally: a bank that deploys a GPT-class model for customer service or document processing, governed only under a general AI policy that never specifically addresses GenAI characteristics (hallucination, prompt injection vulnerability, data extraction risk), is now on notice that general AI governance is insufficient for GenAI.

RBI AI-ACT&RS Compliance Checklist

☐ Board AI risk gap assessment: gap assessment against AI-ACT&RS requirements conducted; board-approved; documented evidence of board deliberation and approval; submitted to RBI by June 30, 2026 deadline (or with documented remediation timeline if deadline passed)
☐ GenAI governance framework: all GenAI tools deployed within the institution inventoried; GenAI-specific governance documentation covering model selection criteria, deployment controls, output validation, human oversight checkpoints, and incident response
☐ AI-driven threat preparedness plan: time-bound action plan for AI-enabled cyber threat detection, response, and resilience improvement; specific measures for frontier AI vulnerability discovery acceleration; implementation timeline defined and board-communicated
☐ Deepfake and AI social engineering controls: supplementary authentication controls for high-value authorizations (beyond voice/video alone); staff awareness training updated for AI-generated phishing and AI-voice cloning; verification protocols for communications purportedly from senior management
☐ Synthetic identity fraud controls: KYC and onboarding AI updated or supplemented to detect AI-generated composite identity documents; biometric verification assessed for AI-spoofing vulnerability
☐ CIS gap assessment: formal assessment of gaps in existing Cyber Information Security protections specific to AI-accelerated threat vectors; gaps documented; remediation integrated into AI threat preparedness plan
☐ 2024 Master Circular integration: AI-ACT&RS obligations mapped to and integrated with existing 2024 Master Circular cybersecurity framework; no gaps where AI-specific cyber threats fall outside either framework's scope
☐ RBI examination readiness: board-approved gap assessment, GenAI governance documentation, threat preparedness plan, CIS gap assessment — packaged for RBI inspection