1. Regulation is defining where AI can be used

Regulators across Asia Pacific are addressing AI within the specific controls that already apply to insurance.

India’s Insurance Regulatory and Development Authority of India (IRDAI) has established an AI working group covering governance, oversight, policyholder protection, and data security. It has also asked insurers to “AI-proof” their systems ahead of Bima Sugam, a digital insurance marketplace intended to connect policyholders, insurers, intermediaries, and other participants. Its guidance highlights bias, privacy, and human oversight in AI applications including underwriting, claims, and fraud detection.

In Singapore, the Monetary Authority of Singapore (MAS) developed an AI Risk Management Toolkit with 24 financial-sector organisations, including insurers, banks, and capital-market firms. The toolkit covers traditional AI, generative AI, and agentic AI, with guidance on governance and risk management.

Other markets are using different regulatory mechanisms. Hong Kong has expanded its GenAI Sandbox++ to insurance, allowing firms to test applications in underwriting, claims, customer experience, risk management, and anti-fraud. Australia’s APRA has focused on governance, risk management, assurance, and operational resilience as insurers adopt AI. Japan’s Financial Services Agency is examining AI-related risks, including cyber threats, while South Korea’s AI Basic Act includes human-oversight and transparency requirements for high-impact AI.

The regulatory question is becoming more specific: what controls are needed when AI is used in an insurance process that affects customers, risk, or financial outcomes? For insurers, AI governance sits alongside existing requirements for model risk, data protection, customer treatment, and operational resilience.

2. Claims offer a clear test for AI automation

Ecosystm research finds that 53% of insurers see automating routine insurance work as AI’s biggest business benefit over the next two years.

Claims provide a strong use case because much of the work involves high volumes of documents, established rules, repetitive checks, and predictable process steps.

Singapore’s insurance sector has reported around 98% of claims being auto-assessed, with roughly 60% processed without human intervention. In India, the National Health Authority and IndiaAI Mission are advancing AI solutions for automated AB-PMJAY claims adjudication, including clinical document analysis and checks for fraud and forged or manipulated documents.

The Philippines is developing a related capability through PhilHealth. Its AI-powered pre-validation system checks diagnoses against laboratory results before claims are paid, helping identify inconsistencies during validation.

Japan provides examples of more complex claims processing. Taiyo Life is deploying a system for around 500,000 claims assessments a year that combines generative AI with a rules engine while retaining explainability. Sompo’s Mysurance and LeadInX are testing a multi-agent claims assistant that uses AI-OCR and large language models to check accident notifications for completeness and consistency.

The same logic applies outside claims. SBI Life Japan has automated after-call work in its contact centre, removing an average of 3.5 minutes of manual processing per call and targeting around 5,000 hours of annual savings.

The common feature is the nature of the work: large volumes of information, defined process steps, and a clear basis for checking the output. These are the conditions under which AI automation has the clearest operational case.

3. AI is altering how underwriting decisions are made

Underwriters already work with large volumes of structured and unstructured information. AI can analyse medical, behavioural, and other risk information at greater speed and scale, and provide recommendations against defined underwriting criteria.

SBI Life in India is using agentic AI to analyse medical histories, laboratory reports, and other risk indicators for life underwriting. The system reviews information against underwriting requirements and supports risk assessment.

In Hong Kong, Prudential’s AI Underwriter gives financial consultants preliminary underwriting indications at the point of sale. These can include potential acceptance, exclusions, loadings, or requests for further information before a formal application is submitted.

Chinese insurers are using behavioural and vehicle data to personalise motor premiums. Driving habits, routes, location, and vehicle characteristics can provide additional inputs for risk assessment and pricing, particularly for electric vehicles where vehicle technology and available data differ from traditional models.

AI recommendations do not remove the insurer’s responsibility for the decision. Models need to be validated, outputs reviewed, and errors or bias identified, particularly when AI influences eligibility, coverage, or premiums. Insurers also need clear rules for when an AI recommendation can be accepted, when human review is required, and who remains accountable for the outcome.

4. AI is changing the fraud equation

AI gives insurers new ways to detect suspicious activity, while giving fraudsters new ways to manipulate the evidence insurers rely on.

South Korea is developing a national AI-based insurance fraud detection system involving financial authorities, health authorities, law enforcement, and insurance organisations. Korea Post Insurance is combining machine-learning models with existing rules-based detection to identify suspicious claims and patterns.

Other national claims systems are building similar checks into the process. PhilHealth is incorporating fraud checks into its claims-validation process, while India’s national health-claims initiative is testing AI to identify forged documents and other manipulated information.

AI is also making fraudulent evidence harder to identify. Deepfakes, synthetic documents, and manipulated identities can be used to support false claims or conceal fraudulent activity. This puts greater pressure on insurers to verify the information behind a claim. AI can help identify patterns and inconsistencies across large volumes of data, but insurers also need controls to assess whether the underlying evidence is genuine.

AI is creating new risks for insurers as well. Autonomous AI agents can cause losses that do not fit established definitions of a cyber incident. MSIG, QBE, and Beazley are reviewing policy language in response, while limited claims history makes these risks difficult to price.

Insurers need to address both sides of the problem: detecting AI-enabled fraud and defining and pricing risks created by AI.

5. New risks create new insurance products

Ecosystm research finds that 47% of insurers expect AI to drive new insurance products, services, and customer propositions.

Some opportunities are directly linked to technologies that create risks insurers need to cover. In Australia, QBE has identified AI, data centres, and clean-energy infrastructure as emerging areas of insurance opportunity, including potential coverage for AI liability and data breaches.

China’s “sci-tech insurance” initiative takes a broader approach, using insurance and risk-sharing mechanisms to support technology companies and emerging technologies. Around USD 1 trillion in coverage was reported in 2025.

Other opportunities come from new ways of structuring insurance itself. IAG’s Firemark Ventures has invested in Adaptive Insurance, whose parametric products trigger payouts when predefined conditions, such as rainfall or wind strength, are met. External data determines the payout rather than a conventional claims assessment.

These opportunities depend on insurers being able to define and price risks where historical claims data may be limited. That is particularly difficult for technologies such as AI, where capabilities and potential failure modes can change faster than insurance experience accumulates.

The commercial opportunity is tied to an underwriting problem: new risks can create demand for insurance before insurers have enough evidence to price them with confidence.

6. The constraints are connected

Ecosystm research finds that insurers face challenges across data, skills, cost, trust, and workflow integration as they scale AI.

These challenges are difficult to address in isolation. Insurance AI depends on sensitive personal, health, financial, and claims data, much of which sits across systems built around established insurance processes. Connecting that data to AI applications requires technical skills, process knowledge, security controls, and people who can oversee the results.

The same dependencies affect the economics. An AI application may reduce manual effort, but the business case also needs to account for data preparation, integration, security, model monitoring, skills, and ongoing technology costs. This makes productivity gains easier to demonstrate than broader financial returns.

Trust also depends on where AI is used. Customers may have little reason to question AI processing a routine back-office task, but expectations are different when it contributes to a premium, underwriting decision, or claim outcome. These applications require stronger controls around explainability, review, and accountability.

The practical challenge is building the conditions for AI to work reliably: usable data, integrated workflows, appropriate skills, a credible return, and controls that support trust.