Conclusion — Leaving No One Alone with Risk

The tools described in this book provide real knowledge. Insurers now have access to larger datasets, more detailed maps, connected devices, climate models and systems that can classify thousands of files quickly. They can estimate some losses more accurately, identify concentrations of risk, adjust reserves and take effective prevention into account. These advances are neither imaginary nor useless [1].

They nevertheless change the insurance promise. The same tool that makes risk more visible can be used to screen risks earlier, differentiate prices more sharply and withdraw before catastrophe. A difference once absorbed by a pool can become a surcharge, a deductible, an exclusion or a refusal. Calculation has not simply become more precise. It now plays a more direct part in deciding who will remain protected and who will be left to carry a growing share of exposure alone [2]. A score may look individual while making it harder for people affected by the same rule to recognise their common position [3].

Climate change makes this tension impossible to ignore. It increases some losses, concentrates them in places that may already be fragile and tests contracts designed for events that were more dispersed [4]. Climate is not the sole cause of the insurance crisis. It reveals what the personalisation of risk had already begun to displace. As exposure becomes more visible, sharing becomes harder to avoid as a political question. Who should pay when a danger is known but the person carrying it can neither reduce it nor leave the place where they live?

This book has not argued for a return to blind pricing. A premium with little relation to expected losses can weaken the insurer, discourage prevention and make others bear a cost that no one has discussed. The issue is whether greater accuracy is being mistaken for a complete answer to a question of justice.

Precision Does Not Decide How Risk Is Shared

A premium is addressed to one person, but it distributes a burden among many. Adding a variable to a pricing rule does more than reveal a difference that was waiting to be found. It gives that difference an economic consequence. A model may show that one group has a higher average cost. It cannot decide by itself whether that cost should be paid by the group, spread more widely, reduced through prevention or covered as a matter of right.

The phrase individual risk can hide how this estimate is produced. Every individual estimate is built from comparisons with other people, data collected from particular contracts and losses recognised through particular claims procedures. People who were refused, those who gave up on a claim and those who paid for repairs themselves are less visible. Even the most personalised price is produced by an institution, a portfolio and a history of classification.

Behavioural data do not solve the problem. They may connect price more closely to an action that genuinely changes danger. They may also turn a constraint into an apparent choice. Driving at night may be required by work. Repairs may be delayed because credit is unavailable. Limited physical activity may reflect disability. Information can be closer to an action without being closer to responsibility.

More information can even reduce the scope for sharing without reducing total loss. As people are classified more finely, those who appear less exposed may refuse to contribute to others. The pool fragments while the danger remains [5]. What remains common is not merely what the model failed to explain. It also reflects what an institution decides not to turn into an individual charge.

Automation does not remove this decision. The language of artificial intelligence can make a tool appear to impose its own logic. We should instead name the task assigned to the system, the information it uses and the consequence attached to its output [6]. Estimating, ranking, recommending and refusing are different operations. They exercise different forms of power.

Risk Remains When the Market Withdraws

A person or place declared uninsurable does not stop being exposed. The possible loss moves elsewhere. It may fall on the household, property owner, municipality, bank, hospital or taxpayer. Withdrawal transfers a burden. It does not make the burden disappear.

This point takes us beyond a simple opposition between market and state. Insurance markets are constituted through public rules, coverage obligations, last resort arrangements, reinsurance and insolvency procedures [7]. Government also intervenes when catastrophe exceeds private reserves or when an essential need cannot depend on an ordinary contract [8]. Private insurance and public protection are not separate worlds. They form an architecture whose division of responsibilities is often difficult to see.

Not every risk should be made insurable. Necessary medical care, a minimum income, emergency housing and evacuation may be direct rights rather than conditional promises. Nor is it always desirable to preserve insurance at any cost. A guarantee that repeatedly finances rebuilding in the same place may maintain exposure. An artificially low premium may delay work and hide a danger that has become persistent.

Information should instead become the starting point for action. A map or score can trigger an assessment, provide finance for building work, create a transition period, support relocation or organise collective prevention. Price alone is inadequate when the person has no means to change the risk. Warning someone without giving them any capacity to act often amounts to announcing their exclusion earlier.

Affordability cannot be separated from the protection that remains. A premium may reflect the risk and still be impossible to pay. A deductible may protect the balance of the contract while making coverage unusable when loss occurs. An insurance architecture must therefore be judged by more than its price. We must ask what protection remains accessible, how much loss stays with the person and what happens when ordinary coverage is no longer available.

A Promise Must Be Enforceable

Insurance turns a contribution paid today into a future claim. Its value becomes visible when loss occurs and the policyholder depends on clauses, evidence, assessment and the organisation’s timetable. A credible promise requires more than financial solvency. It also requires enforceable deadlines, reasoned decisions, access to the file, the possibility of obtaining an independent assessment and an appeal that can be used before the claimant has exhausted their resources [9]. Insurance produces solidarity only through the concrete rules that determine who can claim and under what conditions [10].

The same requirement applies before loss. People must be able to access the data behind a decision, report an error and understand the consequence attached to a category. The presence of a professional who formally approves the result is not enough if that person lacks the time, information or authority to depart from it. Responsibility cannot be divided among the model, vendor, caseworker and final decision maker until no one can be held to account.

Individual appeal remains necessary, but it rarely reaches a rule that produces many similar decisions. Files must be brought together. Aggregate outcomes must be published. Refusals must be documented. Affected groups need a way to intervene before the consequences become irreversible. Counter-expertise then becomes collective. It does not merely ask for a more accurate model. It can challenge the information admitted, the task automated and the right to attach a price or refusal to the result.

An insurance architecture can be assessed by following this entire chain. What problem is it meant to solve? Who enters the pool? How are contributions calculated? What does the guarantee actually pay? Who controls the data and reserves? Who can challenge the rule? Which forms of prevention are financed? What happens to the person left outside? These questions extend existing approaches to public insurance and prevent financial balance from becoming the only measure of success [11].

Following the chain may lead beyond the existing contract. It may require a wider pool, an obligation to offer cover, a public fund, a minimum guarantee, a territorial policy or the removal of some needs from the insurance market altogether. Whatever the solution, a decision about sharing should not be presented as the natural consequence of a model.

Solidarity Is an Institution

Solidarity is not a permanent moral disposition and not the unexplained remainder of a tariff. It depends on rules, resources, rights and organisations able to survive bad years. A mutual insurer, public scheme or territorial fund must also set contributions, choose priorities and decide who may enter. No arrangement eliminates boundaries.

The difference lies in whether those boundaries can be discussed. People affected by them should be able to know how risk was defined, which information was retained, who receives protection and who bears the cost. Ulrich Beck used the term relations of definition for the powers that decide what will count as risk, evidence and a legitimate response [12]. A model is only one part of that chain.

Choice among providers still matters. It does not replace the ability to debate guaranteed needs and acceptable transfers. A person is more than a consumer who can leave a contract. In housing, health, credit and mobility, the exit of some people may weaken protection for others and make the market narrower still [13].

Prevention shows why solidarity must also endure over time. When it works, the house does not burn, hospital admission is avoided and a warning does not become a catastrophe. The expenditure appears in the accounts while the avoided loss remains a counterfactual. Institutions must still finance that absence, preserve a record of the work and prevent the benefit from disappearing when the insurer, contract or owner changes [14].

Elinor Ostrom’s work on resources managed in common remains useful here. Rules must reflect local conditions, responsibilities must be identifiable and avenues of challenge must be accessible [15]. Decisions must also be coordinated across several levels when no single organisation can carry the whole problem [16]. Protection against flood, wildfire or epidemic often extends beyond one property. It requires an institution able to connect private action, common infrastructure and national resources.

Conflict remains, but it has a place. Difficult choices can be discussed before they appear as a series of private refusals, homes that can no longer be sold or treatments that are interrupted. It turns exposure assigned to one person into a problem that several institutions must address.

The Price Must Not Have the Last Word

A risk-priced society is not simply a society that measures a great deal. It is one in which prices and scores increasingly condition access to housing, care, credit, work and territory. The disputes may appear technical. They concern a variable, a map, a threshold or a model. That language can hide a simpler and harder decision. Who should remain in solidarity with whom as danger grows?

Climate change makes the question urgent, but the question extends much further. It arises whenever an institution uses finer information to move each person closer to the cost they are expected to bring. Risk then appears to be a personal property even though its causes and consequences remain largely collective. Infrastructure, work, inherited circumstances, housing policy, access to care and planning decisions continue to shape exposure.

The answer is not to remove numbers. It is to deny them the final word. Models should help anticipate losses, finance promises and guide prevention. They should not decide on their own which vulnerabilities deserve protection, which differences should be charged and who may be abandoned.

We will probably become better at identifying who is exposed, where and when. That knowledge can open rights, finance work and direct resources. It can also give each person an ever more precise explanation of why they will be left to manage alone. The choice between these two uses belongs neither to the algorithm nor to the climate. It belongs to us.

References

1.
Bénéplanc G, Charpentier A, Thourot P. Manuel d’assurance. Paris: Presses Universitaires de France; 2022.
2.
Fourcade M, Healy K. The ordinal society. Cambridge, MA: Harvard University Press; 2024.
3.
4.
Embrechts P, Hofert M, Chavez-Demoulin V. Risk revealed: Cautionary tales, understanding and communication. Cambridge University Press; 2024.
5.
Hirshleifer J. The private and social value of information and the reward to inventive activity. The American Economic Review. 1971;61(4):561–74.
6.
Bender EM, Hanna A. The AI con: How to fight big tech’s hype and create the future we want. New York: Harper; 2025.
7.
Harcourt BE. The illusion of free markets: Punishment and the myth of natural order. Cambridge, MA: Harvard University Press; 2011.
8.
Moss DA. When all else fails: Government as the ultimate risk manager. Cambridge, MA: Harvard University Press; 2002.
9.
Ewald F. L’état providence. Paris: Grasset; 1986.
10.
Lehtonen TK, Liukko J. Producing solidarity, inequality and exclusion through insurance. Res Publica. 2015;
11.
Feinman JM. Designing public solutions for disaster insurance market failures. Journal of Insurance Regulation. 2025;44(2):1–26.
12.
Beck U. World at risk. Cambridge: Polity; 2009.
13.
Self P. Government by the market? The politics of public choice. London: Macmillan; 1993.
14.
Charpentier A. Accounting for prevention. 2026.
15.
Ostrom E. Governing the commons: The evolution of institutions for collective action. Cambridge: Cambridge University Press; 1990.
16.
Ostrom E. Polycentric systems for coping with collective action and global environmental change. Global Environmental Change. 2010;20(4):550–7.