The supply of commercial health insurance is continuously expanding. From the perspective of coverage period, there are both one-year products and long-term products with guaranteed renewal for many years; From the perspective of coverage limit and service level, it covers various types such as million dollar medical insurance, mid to high end medical insurance, etc; In addition, various regions have successively launched city customized commercial medical insurance. At the same time, AI (artificial intelligence) has begun to enter the underwriting, pricing, claims, and health management processes.

Insurance is the process by which policyholders transfer potential losses that may occur in the future and are difficult for individuals to bear to the insurer by paying a predetermined premium. The essence of insurance is a risk transfer mechanism, whose core functions include economic compensation, fund financing, and social management.

Risk sharing refers to the practice of insurers pooling a large number of similar but different risks into a risk pool, allowing the premiums of the majority to compensate for the losses of the minority. It is not about not distinguishing risks, but rather retaining the allocation between groups and time dimensions after risk classification.

In the past, due to limitations in data and technology, insurance companies could only conduct relatively rough risk grouping. The emergence of AI has improved the accuracy of identification and pricing, making individual risk exposure increasingly transparent. This is not only a normal evolution of insurance operations, but also helps alleviate the problem of adverse selection.

But what is truly worth questioning is: as risk classification becomes increasingly precise, should insurance bear more uncertainty, or should individuals, families, and basic medical insurance bear more and more high risks, thus deviating from the original intention of insurance risk sharing?

1、 When AI sees risks clearly, who is blocked from the risk pool

In the past, health insurance underwriting mainly relied on policyholders filling out health questionnaires, followed by manual review of physical examination reports and past medical records. Nowadays, AI is turning this process into automated risk identification: disease history, physical examination indicators, medical records, and claims data are converted into risk labels, and the system can quickly provide results such as standard underwriting, fee increases, liability exclusions, extensions, or refusals.

At the plenary session of the 2026 Lujiazui Forum on "Improving the Effectiveness and Accuracy of Inclusive Finance", Li Yuanxiang, CEO and President of AIA Insurance Holdings Limited, stated that digitalization and artificial intelligence can enable the insurance industry to achieve dynamic pricing and underwriting.

China Ping An's 2025 annual report shows that 94% of life insurance policies achieved instant underwriting that year; AI agent services have been provided approximately 1.702 billion times, accounting for 80% of the total customer service volume; The "AI+manual" system promotes a 30% increase in the number of policy reversals. For insurance companies, this means lower audit costs and higher risk identification efficiency; For consumers in good health, it means faster insurance coverage and simpler process.

Fine identification itself is indisputable, and the essence of underwriting is to determine whether to underwrite and specific conditions based on risk. However, AI may also quickly solidify minor anomalies that previously required comprehensive judgment into rigid risk levels. As a result, people with hypertension, diabetes, nodules and other diseases may face the situation of increased fees, exclusion, derating and even denial of insurance.

This is also the first paradox: AI allows insurance companies to better understand individuals, but may make it harder for the industry to form a sufficiently broad risk pool. Insurance companies should still provide reasons for decisions such as refusal, fee increases, and liability exclusions, retain manual review channels, and offer tiered product plans or transfer options. Otherwise, intelligent underwriting will eventually become an efficient booster for the preference of "only welcoming healthy people".

The problem is no longer just about the accuracy of the algorithm, but also about who will bear the identified risks. If all companies rely on similar data models to compete for low-risk customers and leave high-risk groups to society. The insurance company has only completed risk screening and has not truly achieved risk sharing.

2、 The Paradox of AI Precision Pricing: After Recognizing Risks, Who Will Share Them

Insurance needs to differentiate risks. Factors such as age, medical history, and occupation can all affect insurance premiums. If risk differences are not considered at all and premiums are treated equally, low-risk individuals may withdraw due to high prices, while high-risk individuals may concentrate on entering, ultimately leading to difficulty in sustaining the product.

Insurance is based on the law of large numbers: a large number of people facing uncertain risks pay together, and a small number of people who actually encounter risks receive compensation. But AI can divide the crowd into smaller and finer segments. If everyone pays based solely on their own probability of illness and expected medical expenses, high-risk individuals will need to pay premiums close to future medical expenses, while low-risk individuals will only pay extremely low prices, and insurance will gradually degenerate into "prepaid" medical expenses.

This is precisely the paradox of AI precise pricing: the clearer the risk is seen, the smaller the space for cross population risk sharing may be.

AI can enable insurance companies to achieve more accurate dynamic pricing, which can be roughly divided into two levels: one is to classify risks based on individual health conditions when applying for insurance, forming different conditions such as standard underwriting, additional fees, exclusions, and limits; Secondly, insurance companies adjust the overall product rates based on changes in the payout ratio, medical inflation, and disease incidence rate of the entire product risk pool.

Dynamic pricing should not be understood as consumers being individually raised in price as soon as they fall ill. The regulatory authorities introduced a long-term medical insurance rate adjustment mechanism in 2020, allowing insurance companies to adjust rates based on actual operating conditions, but the first adjustment must not be earlier than three years after the product is launched, and the interval between two adjustments must not be less than one year. The basic logic is to calibrate the overall price of a product risk pool, rather than repricing it for a specific patient.

The positive value of AI is to refine the simple "underwriting or denying insurance" into fee increases, limits, and exclusive products, providing opportunities for some chronic disease and non-standard body populations to enter the risk pool. In 2026, Pacific Health Insurance, Ant Health, and China Re Life Insurance will launch the "Blue Health Insurance · Exempt Health Report Long Term Medical Insurance (Lucky Edition)", which includes general past medical conditions as protection and uses AI intelligent agents to provide report interpretation, health consultation, and risk management services.

This represents another use of AI: not only to identify risks before insurance, but also to reduce risks through disease prevention, chronic disease management, and medical services after underwriting. If insurance companies can reduce unnecessary medical expenses, they may use the saved costs to expand coverage.

AI can make risks transparent, but it cannot make them disappear out of thin air. The more accurately insurance companies can identify high-risk individuals, the more the industry needs to answer who ultimately bears these risks. Accurate pricing should not end with risk elimination, but should become the starting point for tiered underwriting, health management, and expanded protection. Otherwise, the algorithm enhances the screening ability of insurance companies, but weakens the fundamental risk sharing ability of the insurance system.

Accurate pricing is not a sin, the boundary lies in whether insurance companies are looking for ways to absorb risks or pushing risks back to individuals.

3、 How to rebuild a risk community in the era of AI

A more scientific risk sharing approach does not require everyone to pay the same premium or purchase the same product, but rather provides clear and interconnected protection based on the risks, needs, and payment capabilities of different groups of people.

Young, healthy and price sensitive people can choose one-year Internet medical insurance, but they need to fully understand the risks of renewal and termination; People who value long-term stability can choose insurance renewal products and accept reasonable price adjustments based on changes in overall medical costs; Patients with chronic diseases, the elderly, and those with pre-existing conditions need to take on sick body insurance, tax premium health insurance, and urban universal insurance; People with higher payment ability and medical service needs can choose mid to high end medical insurance.

Shanghai Huibao "is a sample of multiple companies sharing risks together. This project is underwritten by China Taiping Life Insurance, with participation from 8 insurance companies including China Life, Xinhua Insurance, PICC Health, Taikang Pension, Ping An Health, Taiping Pension, and CCB Life Insurance. The products for the year 2026 are not limited by age, occupation, or health status. Since its launch in 2021, it has covered over 33 million people and compensated over 3 billion yuan.

But universal benefit does not mean unlimited responsibility. In 2026, the "Shanghai Huibao" policy will uniformly adjust the reimbursement ratio for past medical conditions related to hospitalization self payment and CAR-T related responsibilities to 30%, and set annual compensation limits for some drugs. This indicates that even if a co insurance model is adopted, a balance needs to be found between inclusiveness and sustainability.

The industry also needs to diversify the pressure of compensation through co insurance, reinsurance, and compensation mechanisms for high-risk groups, in order to avoid excessive concentration of risk borne by individual companies due to expanding underwriting. Under the premise of legal authorization and privacy protection, it is also necessary to promote the sharing of necessary data among medical insurance, hospitals, and commercial insurance, so that AI can be used more for medical expense analysis, disease prevention, and abnormal claims recognition, rather than simply screening out high-risk individuals.

In June 2026, the State Administration for Financial Regulation issued guidance on the safe development and application of artificial intelligence in the banking and insurance industries, requiring financial institutions to promote AI application compliance, transparency, and trustworthiness, and to develop in a beneficial, safe, and fair direction. For health insurance, this means that data usage should have boundaries, algorithm results should be interpretable, and models need to regularly check for implicit discrimination against specific diseases, occupations, regions, or income groups.

To measure the value of AI applications, we should not only look at how much underwriting speed has increased or how much labor costs have decreased, but also at whether it has reduced premiums, improved underwriting conditions, and provided protection for more high-risk populations.

Conclusion

What health insurance really sells is not a simple contract that a person can easily buy when they are healthy, but the qualification to remain in the risk community even after their health condition changes.

AI can assist insurance companies in underwriting faster, pricing more accurately, and can also be used for disease warning, health intervention, and medical expense management. Technology itself does not necessarily hollow out risk sharing. The key lies in what insurance companies choose to use it for: to more accurately screen out high-risk individuals, or to more effectively manage risks and provide tiered protection for those who were previously unable to purchase insurance?

Commercial insurance requires scientific pricing and reasonable profits, and products with short-term, long-term, and different risk levels all have value. However, precise pricing cannot lead to precise exclusion, dynamic price adjustments cannot exceed the guarantee of renewal commitment, and algorithms cannot be a reason for insurance companies to refuse coverage.

AI can divide each person's risk into smaller and smaller segments, while the insurance system needs to reorganize these segmented individuals into different levels of risk pools. The more technology can see risks clearly, the more insurance cannot only see risks. Only when AI is used more to expand coverage, manage medical expenses, and improve health can health insurance maintain the core of "risk sharing" in the wave of intelligence.