Introduction
Insurance is a term used to describe activities of organisms that have the right to engage in covering everyday life risks that might not happen, and if they do, it is not certain when and for how much (Marly, 2024). It is a service sought out with the intention of being covered in the event of a disaster, such as a traffic accident, sudden death, housefire, but with hopes that it will never be used.
As any business, insurers seek profit in a competitive and fast-paced market. This pushes them to develop different ways of achieving optimal and dynamic pricing for their services, charging their clients in a fair way with the aid of new technologies, such as artificial intelligence (AI) and machine learning (ML).
Although highly efficient for gathering and analysing objectively large sums of data, the AI does not take into consideration individual traits, as they are also not trained for doing it nor is it possible due to the amount of data that needs to be treated constantly. AI does not take into consideration the fact that Joe, who is 23 years old, never had an accident even though he is part of a social group considered as high risk of getting involved in traffic accidents (those under 25 years old), whereas Bob, who is 45 years old and theoretically should be more experienced when driving, has an accident every year. Joe is still going to pay a higher premium for his automobile insurance than Bob.
The techniques used by insurances to gather and assess the data that will influence the price of the services can lead to mass generalisation which can end up excluding or discriminating certain groups and certain people within those groups, increasing the costs of having an insurance based on protected characteristics, such as gender, age, disabilities, and ethnicity.
Assessing the risk in insurance contracts
Insurers offer important services to modern societies, such as motor insurance aiding people’s mobility, household insurance protecting property, and life insurance for protecting family members against poverty caused by sudden invalidation or death of the provider. However, those same companies also aim for profit, as they are businesses and are a part of the economic market, such as any other company.
Like any private service, customers must pay a fee to the provider in order to have their risks covered. Insurers determine the premium by assessing factors such as the vehicle’s age and weight, the customer’s age or medical history, and the location of the property to be insured. These elements help them estimate the level of risk and the likelihood of the client filing a claim, using this knowledge as source to create a premium.
Insurances have been using AI to analyse their data for several years now, the most common way being the data-intensive underwriting, when the intelligence gathers and studies a large sum of data at the same time to estimate the percentage of risk in predetermined social groups and, therefore, stipulate the premium for the customers that are part of it (Bekkum et al, 2025).
Another usage of AI in data analysis is through the behaviour-based risk assessment, when insurers base their premiums on the behaviour of each customer, individualising the risk. An example of this would be a car insurer that offers discounts to consumers that drive safely according to a device in their car, showing that there is less risk of the disaster happening and the insurer having to cover for it. Or a health-policy that bases its costs on a client’s health data collected through his smart watch or a phone application, producing more dynamic information. It is a shift in the insurance market that is still at the beginning but shows signs of rapidly growing (Vinichuk, 2025).
Commonly those companies assess the risk by analysing group characteristics and quantifying them to achieve percentages of how many disasters happen in a determined group, in a determined location, and so on. Instead of analysing if an individual drives safely, the insurance will assess how frequently accidents happen in certain groups, such as people under 25 years old or those who drive heavier vehicles and then design the cost of the service according to it. It is a matter of ex-ante profiling, when personal information is assessed to create a profile definition that might correspond to a specific level of risk based on a probabilistic analysis, in general terms.
This profiling is used to justify charging different insurance premiums to different groups. However, it is often perceived as discriminatory toward individuals within the profiled group who have not engaged in the behaviours that would warrant a higher premium. In such cases, the additional amount they pay brings no corresponding cost to the company, making these individuals more profitable. In other instances, they may even be excluded altogether despite not engaging in the risk-increasing behaviours, as a result of discriminatory decisions made by the companies.
Direct and Indirect Discrimination
The European Union’s non-discrimination directives together (2000; 2004; 2006; 2008) prohibit discrimination for six protected characteristics: age, disability, gender, religion, racial or ethnic origin, and sexual orientation. However, not all the directives apply to all situations, some of them targeting specific audiences and situations, such as those that focus exclusively on employment context.
When it comes to insurance, the directives only prohibit discrimination based on gender and ethnicity, meaning that many other forms of discrimination that might be committed by insurers are not illegal under the directives, as they do not harm groups mentioned before, but that does not mean that they are not equally unfair.
As private entities, insurers are largely governed by contractual freedom, meaning that they can choose with whom they want to sign a contract, for what price, and under which conditions. The Directives will act upon this freedom to forbid blatant discrimination against the aforementioned group, although their market freedom is still large.
In Europe, two types of discrimination are prohibited: direct and indirect. Direct discrimination happens when an organisation differentiates people based on a protected characteristic, such as ethnicity or gender. An example would be the differentiation based on ethnicity that insurers in the United States (US) used to do in the 1950s, claiming that African Americans, on average, died sooner than white Americans and, therefore, should pay a higher cost when contracting health or life insurance.
Insurances cannot discriminate against people based on ethnicity or gender, as controlled by the Directives. This prohibition is backed up by the 2011 case of the Test-Achats (Test-Achats ASBL and Others (C-236/09) ECLI:EU:C:2011:100), when the Court of Justice of the European Union (CJEU) prohibited life insurers of using sex as a differentiator in the assessment of insurance risks. The judgement was based on articles 21 and 23 of the Charter of Fundamental Rights of the European Union (CFREU), which prohibit any discrimination on grounds of sex and require equality between men and women to be ensured in all areas of society.
Another example of direct indiscrimination is a case from 2018, in the Netherlands, when an insurer denied home insurance to those living in caravans, which was judged as a direct discrimination against caravan dweller people who were excluded from benefiting from the service. By unilaterally terminating their coverage, the insurance company was violating the principle of non-discrimination against ethnic groups.
Additionally, it is important to note that any discrimination based on sex, gender, health state, genetic characteristics or gender identity can be criminally punished by EU state members.
On the other hand, indirect discrimination occurs when a practice is initially considered as neutral but ends up discriminating against the protected people. An example would be a home insurer that charges higher premiums for homeowners located in a certain neighbourhood, disregarding the fact that this location mostly includes people of a certain ethnicity. It was not the plan of the insurer to discriminate against that group of people, but it ended up happening when it decided to raise the premium for that specific area. This shows that data-intensive underwriting can unintentionally lead to discrimination, even if it was not clear at the beginning.
Avraham and D. Minty (2021) believe that “the insurer must choose the least intrusive form of differentiation and must ensure that a practice does not cause disproportionate disadvantages for protected groups”, leaving them in a more precarious position than before.
Another example would be an insurer that offered life insurance over the internet and decided to charge higher prices to those most likely to die earlier, which often meant the poorer people, as poverty comes attached to many factors that contribute to shorter lifespans, such as malnutrition, lack of access to healthcare, exhausting work for long periods of time, and so on. This insurer would perpetuate the economic inequalities by making it harder and more expensive for these people to access a service that protects their family from imminent economic misery in the event of a sudden death.
In a legal aspect, in this case the EU non-discrimination law does not prohibit the differentiation from being made unless it also affects people of a certain ethnicity or gender, as they are part of the protected characteristic.
The Finance Watch’s Policy Brief (2019) shed a light on the fact that the right of European citizens to equal treatment is currently not being protected under financial services law, calling for the importance of guaranteeing citizens full and equal access to the basic kinds of insurance. It states that insurance premiums based on general behavioural dimensions are discriminative when they are not strictly calculated based on actual damages incurred.
Personalised behaviour-based insurance could lead to more inclusion as a better understanding of the risks can improve financial inclusion for high-risk consumers who would not be able to afford insurance otherwise, as stated by the European Insurance and Occupational Pensions Authority (EIOPA). Also, it would not be viewed as discriminatory if the distinction is used to create discounts to customers with a clear and objective reasoning behind it, instead of increasing the costs of the premium for others.
It is necessary that insurers adopt different methods to assess the risk that can directly impact the pricing of their premium, such as using methods based on the actual claim’s history and historical cost of damages incurred by individual customers, with the aid of AI, for example, individualising the risk and, therefore, avoiding discrimination.
Conclusion
Profiling customers is the traditional way that insurance companies found to put a price on the products they sell. It consists of dividing potential consumers and allowing the same companies to define their target markets. By grouping a significant amount of data and analysing it via AI or ML, insurances stipulate the cost that each customer will have to pay as premium depending on factors such as age, health, size and age of the vehicle, neighbourhood where the house is placed, that are analysed in a large pool of data collected throughout the years.
By taking into consideration only general statistics in a data-intensive risk assessment, insurances can discriminate directly or indirectly against customers that end up paying higher premiums or being forbidden from buying cover due to their protected characteristics, leaving them out of the necessary protection.
Companies must shift their perspective and adopt new methods of assessing customer risk to set premiums fairly, such as individual behaviour-based insurance and personal history, ensuring equal access to insurance while complying with EU directives that safeguard the right to equal and fair treatment.
Bibliography
Legislations
Charter of Fundamental Rights of the European Union, OJ 2012 C 326/391. Retrieved September 23, 2025 from: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:12012P/TXT.
Court of Justice of the European Union (Grand Chamber), Test-Achats ASBL and Others v Conseil des ministres (Case C-236/09, Judgment of the Court of 1 March 2011) ECLI:EU:C:2011:100. Retrieved September 20, 2025 from: https://curia.europa.eu/juris/document/document.jsf;jsessionid=2CC58A7CB9CB70D2B48D231D43067F79?text=&docid=80019&pageIndex=0&doclang=en&mode=lst&dir=&occ=first&part=1&cid=13406548.
Code pénal [French Penal Code] art 225-1 and 225-2, Légifrance, version en vigueur au 23 septembre 2025. Retrieved September 17, 2025 from : https://www.legifrance.gouv.fr/codes/section_lc/LEGITEXT000006070719/LEGISCTA000006165298/?anchor=LEGIARTI000045391831#LEGIARTI000045391831.
Directives (EU) No 2000/43/EC; 2000/78/EC; 2006/54/C; 2004/113/EC and Directive Proposal (COM (2008) 462). Non-Discrimination. Retrieved September 20, 2025 from: https://commission.europa.eu/aid-development-cooperation-fundamental-rights/your-fundamental-rights-eu/know-your-rights/equality/non-discrimination_en.
NJCM, Reaal Woonverzekeringen <https://oordelen.mensenrechten.nl/oordeel/2018-14>. Retrieved September 23, 2025 from: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:62012CJ0267
Regulation (EU) No 1094/2010 of the European Parliament and of the Council of 24 November 2010 establishing a European Supervisory Authority (European Insurance and Occupational Pensions Authority), amending Decision No 716/2009/EC and repealing Commission Decision 2009/79/EC, OJ L 331, 15.12.2010, p. 48-83. Retrieved September 19, 2025 from: https://eur-lex.europa.eu/EN/legal-content/summary/european-insurance-and-occupational-pensions-authority-eiopa.html.
Online sources
Bekkum, M. V. (March 11, 2025). AI, Insurance, Discrimination and Unfair Differentiation: An Overview and Research Agenda. Taylor & Fracis Online. Retrieved September 22 from https://www.tandfonline.com/doi/full/10.1080/17579961.2025.2469348#abstract
Cather, D. A. (October 26, 2023). Addressing insurance price discrimination in an era of diversity, equity, and inclusion. Wiley Online Library. Retrieved September 22 from https://onlinelibrary.wiley.com/doi/full/10.1111/rmir.12249.
Guidance Resource: AI and Discrimination in Insurance. Australian Human Rights Commission. (December 1, 2022). Retrieved September 17 from https://humanrights.gov.au/our-work/technology-and-human-rights/publications/guidance-resource-ai-and-discrimination-insurance.
How to Find the Best Insurance Pricing Strategy for Optimal Results. Earnix (May 27, 2024). Retrieved September 18, 2025 from https://earnix.com/blog/determining-the-best-insurance-pricing-strategy/.
Jérusalmy O. and Fox P. (February 25, 2019). Insurance and Discrimination: Guaranteeing Full and Equal Access to Basic Insurance. Finance Watch. Retrieved September 20, 2025 from https://www.finance-watch.org/policy-portal/retail-inclusion/insurance-and-discrimination-guaranteeing-full-and-equal-access-to-basic-insurance/.
Jaap, A. et al (August 2008). Better Safe than Sorry? Ex Ante and Ex Post Moral Hazard in Dynamic Insurance Data. Tinbergen Institute. Retrieved September 21, 2025 from https://research.tilburguniversity.edu/en/publications/better-safe-than-sorry-ex-ante-and-ex-post-moral-hazard-in-dynami.
Kagan, J. (August 22, 2025). Understanding Insurance Premiums: Definitions, Calculations, and Types. Investopedia. Retrieved September 15, 2025 from https://www.investopedia.com/terms/i/insurance-premium.asp.
Minty D. (March 3, 2021). Why Equality of Fairness Will Shape the Future of Insurance. Ethics and Insurance Info. Retrieved September 14, 2025 from https://www.ethicsandinsurance.info/equality-of-fairness/.
Vinichuk, O. (May 8, 2025) Behavior-Based Insurance To Reshape Life & Health Plans. Or Will It? Science Soft Finance. Retrieved September 22, 2025 from: https://www.scnsoft.com/finance/behavior-based-insurance-in-life-and-health-products.
