India's insurance sector is one of the most AI-intensive regulated industries in the country. Motor claims are being assessed by computer vision within seconds of an accident photo being uploaded. Health fraud detection AI is flagging suspicious claims patterns before payment. Life insurance underwriting models are incorporating behavioral data, health wearable signals, and digital footprint. Customer-facing LLMs are handling policy queries at scale. The AI is already there — the governance framework is catching up.
IRDAI's June 2026 working group announcement puts the insurance sector on notice: India's first formal AI governance framework for insurance is being built now, by a seven-member expert group with a three-month timeline. The working group's scope — AI inventory, ethical AI, explainability, cybersecurity risk, stress testing, audit frameworks, capacity building — covers exactly the operational areas where Indian insurers' current AI governance has gaps.
The question for every Indian insurer is not whether IRDAI will regulate AI — that question was answered on June 19, 2026. The question is whether the insurer will arrive at the first examination under the new framework with governance already in place, or with governance that was assembled the week before the examination. The governance principles the working group will codify are already knowable: they will align with RBI FREE-AI, DPDP Act obligations, and the global standards the working group is tasked to benchmark. An insurer that starts governance work now will have a documentation advantage that cannot be created retroactively.
IRDAI Working Group Scope: What the Framework Will Cover
AI System Inventory
The working group will assess AI adoption and governance maturity across the sector and recommend an inventory framework for AI systems in use. This mirrors RBI's model inventory requirement and DPDP Act sub-processor documentation obligations. For insurers, this means identifying all AI systems — vendor-provided and internally built — that affect policy pricing, underwriting, claims assessment, fraud detection, customer communication, and compliance. The inventory requirement extends to the hyperscaler and foundation model layer, following the concentration risk framework that RBI has been developing for banks.
Ethical and Explainable AI
The working group will develop ethical and explainable AI frameworks for the insurance sector, aligned with policyholders' data protection obligations under DPDP Act. The fairness dimension is particularly material in Indian insurance: AI underwriting models that proxy health status through geographic location, device type, or digital behavior patterns could produce discriminatory outcomes against rural, elderly, or lower-income policyholders — the same populations that formal insurance is meant to serve. Explainability for adverse underwriting or claims decisions is both a DPDP Act right (human review of automated decisions) and a policyholder fair treatment obligation under IRDAI's own consumer protection framework.
AI-Driven Cybersecurity Risks
The inclusion of CERT-In in the working group signals that the framework will specifically address AI-enabled cybersecurity risks in insurance. This includes: insurance AI systems as targets for adversarial manipulation (manipulating claim photo assessments, poisoning underwriting models, prompt-injecting customer-facing LLMs); insurance AI systems as amplifiers of cyber risk (using AI to facilitate fraudulent claims, identity theft, policy fraud); and insurance AI infrastructure concentration risk (the same hyperscaler concentration risk concerns that apply to banking AI apply to insurance AI). The CERT-In involvement is the most significant signal that the framework will have a security dimension beyond what insurers currently govern.
AI Audit Framework
The working group will recommend an audit framework for AI in insurance — the equivalent of the MRM validation requirement for banks. For insurers, this will likely translate into: pre-deployment validation evidence requirements for AI systems affecting policyholder decisions; ongoing monitoring requirements for production AI; and evidence documentation standards for regulatory examination. Insurers that have been deploying AI without a formal validation program will need to retrofit governance documentation — which is harder and more expensive than building it from the start. The audit framework recommendations are expected to align closely with RBI FREE-AI's accountability and governance pillars.
The IRDAI working group's three-month timeline from the June 19, 2026 announcement puts the framework recommendations in late September 2026 — with formal implementation guidance likely in early 2027. Insurers that use the period between now and formal implementation to document existing AI systems, conduct gap assessments, and build validation evidence for their highest-risk AI deployments will be materially better positioned than those that wait for the formal standard.
AI Use Cases Requiring Governance Now: Motor, Health, Life, and Customer LLMs
Indian insurers are deploying AI at scale in four major functional areas, each with distinct governance obligations under current frameworks (DPDP Act 2023, IRDAI consumer protection guidelines) and emerging IRDAI AI governance requirements:
- Motor claims — agentic AI and computer vision: Some Indian insurers are deploying agentic AI systems that autonomously assess accident photos, verify policy details, check No-Claim Bonus status, and initiate claims settlement without human review at each step. Computer vision has reduced vehicle inspection time by up to 75% in some deployments, and early agentic AI deployments are approaching same-day settlement for straightforward claims. The governance requirement: what is the authorization scope of the agentic system (what actions can it take without human approval?), what is the human oversight threshold (at what claim value or complexity does a human review?), and what is the appeals process when a policyholder disputes the AI's damage assessment? The DPDP Act right to human review applies to AI-generated adverse outcomes, including disputed claim valuations.
- Health insurance fraud detection: AI systems flagging suspicious claims patterns, hospital billing anomalies, and policyholder behavior as potentially fraudulent are deployed widely in India's health insurance market. The governance concern is false positives: a legitimate health claim that is AI-flagged for fraud delays payment to a policyholder in a medical situation, with potentially life-affecting consequences. The governance requirements include: false positive rate monitoring with human review of all flagged claims before payment denial, appeals processes with defined timelines, and audit trails showing the AI's reasoning for each flag (DPDP Act data access obligation). Insurers that deploy fraud detection AI without adequate false positive governance are exposed to both consumer protection enforcement and DPDP Act complaints.
- Life insurance underwriting: Predictive models for mortality risk, persistency (whether a policyholder will lapse), and premium pricing are well-established in Indian life insurance. Governance obligations include: model validation against actuarial standards (the Insurance Institute of India's actuarial validation requirements apply), fairness testing (does the model produce materially different pricing for otherwise-similar risk profiles based on demographic characteristics that IRDAI's non-discrimination principles would prohibit?), and documentation sufficient for IRDAI examination. The FREE-AI advisory recommendation that regulated entities conduct pre-deployment disparate impact analysis applies to life insurance underwriting models under IRDAI's equivalent obligation to treat policyholders fairly.
- Customer-facing LLMs for policy and claims: LLMs deployed in insurance customer service for policy information, claims status, renewal guidance, and product queries are exposing insurers to mis-selling risk at scale. A customer who receives LLM-generated information about policy coverage terms and relies on it in a claims situation, only to find the LLM was incorrect, has a potential claim against the insurer for mis-selling or mis-representation. The governance requirements: LLM accuracy validation for product-specific queries, hallucination rate below threshold for policy coverage terms, clear customer communication that the LLM is an AI assistant and not a binding commitment, and escalation paths to human advisors for coverage-commitment questions.