India’s insurance regulator is close to receiving the sector’s first formal recommendations on artificial intelligence. IRDAI set up a seven-member working group on 18 June 2026, chaired by Sandeep Shukla of IIIT Hyderabad, with three months to report to the member for finance and investment. Its brief covers claims and fraud prevention by name, and asks for an audit framework spanning both pre-deployment and post-deployment checks.
Insurers have not waited for it. Writing in Insurance Asia, Sachin Dutta argues that agentic systems differ from earlier automation because they reason, plan and coordinate actions across multiple systems with limited human intervention. That lets them orchestrate work across underwriting, policy administration, customer servicing, medical assessment and claims management.
Where the agents are actually going to work
Customer service came first. At the ETCIO Annual Conclave 2026, Mukul Jain, chief technology officer at Axis Max Life Insurance, described a multi-agent setup for policy queries in which one model drafts a response and a second validates it before the customer sees anything. He called human-in-the-loop an operating model for the transition rather than a weakness, and said regulated sectors need clear boundaries around where autonomous systems act alone.
Pre-authorisation is under different pressure. The IRDAI Master Circular on Health Insurance Business of 29 May 2024 requires an insurer to decide a cashless authorisation request within one hour, to offer pre-authorisation digitally, and to clear final discharge authorisation within three hours. Where discharge runs past three hours, any extra hospital charge comes out of the insurer’s shareholders’ fund.
Not all of it qualifies as agentic. Bharani Subramaniam of Thoughtworks told the same panel that many systems described that way remain deterministic workflows, and that genuine agentic behaviour appears where the steps are not defined in advance. Himanshu Pant, chief data officer at Adani Group, put the risk plainly: if the processes are not right, AI will only accelerate the error.
What supervisors are asking for
The India AI Governance Guidelines, published by a MeitY drafting committee and summarised by PIB on 15 February 2026, set out seven principles.
- Accountability is assigned by function performed, risk of harm and due diligence conditions.
- Human oversight is to be mandated in sensitive and critical sectors, specifically to limit loss-of-control risk.
IRDAI appears in that document as one of the sectoral regulators expected to enforce domain-specific norms.
Questions worth asking when a decision looks automated
A rejected health claim already carries human accountability by rule. Clause 17 of the Master Circular says no claim may be repudiated without approval from the Product Management Committee or a three-member Claims Review Committee drawn from it. Where a claim is rejected or partly disallowed, the insurer must convey full details citing the specific terms and conditions relied on. So the two questions that matter are:
- Which committee approved the decision?
- Which policy clause does it rest on?
A refusal that cannot answer either is not a compliant refusal, whatever produced it.
Two more points are worth holding onto. The insurer’s reply to any grievance must carry the contact details of the relevant insurance ombudsman. Once an ombudsman issues an award, the insurer has 30 days to comply. On missing that, a penalty of ₹5,000 a day runs in favour of the complainant, on top of penal interest under the Insurance Ombudsman Rules, 2017. Reading a rejection letter against the policy schedule is where most of this gets decided, and it is the part automation does not do for the policyholder. MyRupia works through claim rejections on that basis, holding no insurer tie-ups and taking nothing on the outcome.
Dutta’s own conclusion is that the lasting measure of these systems will be the confidence they earn from customers, employees and regulators.
Disclaimer: This MyRupia article is for informational purposes only and is based on publicly available government, regulatory and industry sources. It should not be treated as investment, financial, tax, insurance, or legal advice. Information, examples, market data, and expert views mentioned in the article may change over time and should not be considered a recommendation to buy, sell, invest in, or surrender any financial product. Readers should evaluate their individual circumstances and consult a qualified financial professional before making decisions.