Actuary

The European insurance supervisor's own Opinion of 6 August 2025, addressed to national competent authorities rather than directly to firms. Read in the full PDF. Paragraph 2.4: the AI Act identifies as high-risk the use of AI systems for risk assessment and pricing in relation to natural persons in life and health insurance. In its description of roles, it states that the actuarial function is responsible for the controls on AI systems that fall under its responsibilities, giving as examples the coordination of the technical provisions calculation and the opinion on the overall underwriting policy. Its annex on fairness metrics warns that some group fairness metrics could contradict actuarial fairness, where customers bearing the same risk are charged the same price. It establishes where responsibility sits in EU insurers; it does not measure how much of actuarial work software does, and it counts no one.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The regulator's own circular letter of 11 July 2024, addressed to insurers authorised in New York, fraternal societies, HMOs and the State Insurance Fund. Its verbs are expectations (should), not a statute, and it covers underwriting and pricing only. It expects unfair or unlawful discrimination testing before an AI system goes into production and on a regular cadence thereafter, lists quantitative methods including the adverse impact ratio, denial odds ratios and marginal effects, expects insurers to show that external data are supported by generally accepted actuarial standards of practice, requires comprehensive documentation, places oversight on the board and senior management, and says insurers retain responsibility for third-party vendor tools. It does not say who inside an insurer does the testing, and it measures nothing about actuaries' work or numbers.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

If you are starting out, the data and first-draft analysis that used to fill a junior actuary's week is the part being tooled fastest. The route that is widening is the one regulators are writing: testing models for unfair discrimination, documenting them, and explaining them to people who must sign. Learning to do that well is the entry point now, alongside the exams.

Your leverage is the signature and the judgement behind it. Regulators are naming the actuarial function as responsible for the controls on AI within its remit, so the actuaries who can read, test and challenge a machine-learning model — not only build a traditional one — are the ones those responsibilities will land on.

Reshape the role

Own the model-testing duty

Regulators now expect pricing and underwriting models to be tested for unfair discrimination before and after they go live. Someone at every regulated insurer has to do that, and an actuary who can do it is combining the regulation, the model and the pricing logic in one person.

Stay and strengthen

Stay where a signature is required

Reserving opinions and the actuarial function's statements are duties placed on a named function by regulation. The work beneath them is being tooled; the responsibility is not moving, and it becomes more demanding as more of the calculation is automated.

Adjacent move

Move toward model risk and AI governance

The regulators on this page describe governance frameworks that need people who understand both the statistics and the insurance: documentation, validation, third-party model oversight. Actuarial training is unusually close to that, and the demand is created by regulation rather than by a cycle.

Not the part regulators hold someone responsible for. The data preparation and first-draft analysis that fill much of a junior actuary's time are being tooled quickly. But regulators are going the other way on responsibility: the European supervisor names the actuarial function as responsible for the controls on AI within its remit, and New York expects insurers to test AI pricing models for unfair discrimination and to show their data meets accepted actuarial standards. So the tasks beneath the signature are being automated while the signature and its duties are growing. What is not known is how many people that combination employs.

We do not answer that with a number of years, and for this occupation there is a better signal you can read yourself: what your regulator asks of the actuarial function. Read your regulator's latest guidance on AI in insurance and count the duties it places on actuaries or the actuarial function. If the list of named duties is growing while the tooling beneath them improves, the occupation is being narrowed toward judgement and responsibility, not removed. If a regulator ever stops requiring a named person's opinion, that is the change to watch for, and it would be published.

In the European Union, AI used for risk assessment and pricing of natural persons in life and health insurance is high-risk under the AI Act, and the insurance supervisor's opinion makes the actuarial function responsible for the controls on AI systems within its remit, for example in coordinating technical provisions. In New York, the insurance regulator expects AI and external data used in underwriting and pricing to be tested for unfair discrimination before production and regularly afterwards, and the data to be supported by accepted actuarial standards of practice. Neither says software cannot be used; both say someone must be able to justify and control it.

Yes, and the European supervisor says so itself: some group fairness metrics could contradict actuarial fairness, where customers bearing the same risk are charged the same price. That is why testing a pricing model is not a mechanical step. Someone has to decide which notion of fairness applies to which product, justify it to a regulator, and document it — and that is a judgement actuaries are trained for and software is not.