Will AI replace actuaries?

No. Actuarial opinions, reserve certifications, and rate filings require a credentialed actuary's signature and professional accountability. AI cannot sign a Statement of Actuarial Opinion or appear before regulators. Under ASOPs 56, 23, and 12, the actuary remains professionally responsible for any model output used in actuarial work, including AI-generated output. AI makes actuaries more productive by automating data processing, but judgment, interpretation, and regulatory communication remain human responsibilities.

How is AI changing actuarial work in 2026?

AI affects actuarial work in three primary ways: actuaries validate AI models used in underwriting decisions under ASOP 56; actuaries certify that AI-driven rate filings are actuarially justified and non-discriminatory under ASOP 12; and AI generates new data streams (telematics, satellite imagery, IoT sensors, natural language processing on claims notes) that actuaries integrate into pricing and reserving models. The NAIC launched a 12-state AI evaluation tool pilot in March 2026 to create a structured framework for regulatory review of insurer AI systems.

What skills do actuaries need for AI work?

Python proficiency is the baseline: enough to read code, run models, and evaluate data science output. Model governance knowledge is increasingly valuable as companies build AI governance frameworks. Understanding of ASOP 56 (modeling), ASOP 23 (data quality), and ASOP 12 (risk classification) is essential for actuaries who sign rate filings or validate models used in underwriting decisions. Communication skills are critical for translating AI model performance into regulatory language and business decisions.

What is the NAIC AI evaluation tool pilot?

The NAIC launched its AI Systems Evaluation Tool pilot across 12 states on March 2, 2026. The participating states are Colorado, Maryland, Louisiana, Virginia, Connecticut, Pennsylvania, Wisconsin, Florida, Rhode Island, Iowa, Vermont, and California. The tool has four exhibits covering AI usage quantification, governance risk assessment, details on high-risk AI systems, and AI data details. The pilot runs through September 2026, with updates and potential formal adoption expected at the NAIC fall 2026 national meeting.

AI Underwriting Adoption Expected to Jump from 14% to 70%: What It Means for Actuaries

Accenture surveyed 430 senior insurance underwriting executives across 11 countries. The NAIC launched a 12-state AI evaluation tool pilot in March 2026. Existing ASOPs already define the actuary's validation and certification responsibilities. This is not replacing actuaries; it is redefining where their judgment is required.

Accenture published a survey of 430 senior insurance underwriting executives across life, commercial P&C, and personal P&C in 11 countries. The headline: AI adoption in underwriting is expected to jump from 14 percent today to 70 percent within three years. That is not incremental change. It is a fundamental reshaping of how insurance risk gets assessed and priced.

The broader market signal is consistent. Separately, the NAIC surveyed U.S. insurers on their AI and machine learning use and found that 88 percent of auto insurers reported they use, plan to use, or plan to explore AI or ML models in their operations. For health insurers, the figure was 92 percent. For home insurers it was 70 percent. These are not pilot numbers anymore. They are mainstream adoption figures.

From tracking regulatory and actuarial standard developments since the first NAIC AI guidance in 2023, the pattern is now clear: AI in underwriting creates categories of actuarial work that did not exist a decade ago, places existing responsibilities under more scrutiny, and rewards actuaries who develop technical fluency alongside their credentialing. The actuaries who understand what is actually changing are better positioned than those reacting to headlines.

14% to 70%

Expected AI adoption in underwriting within three years (Accenture, 430 executives, 11 countries)

88%

Auto insurers who use, plan to use, or plan to explore AI/ML (NAIC survey, 193 auto insurers)

12 states

Participating in the NAIC AI Systems Evaluation Tool pilot launched March 2, 2026

The Regulatory Framework Taking Shape in 2026

The regulatory environment for AI in insurance has moved from principles to enforcement infrastructure between 2023 and 2026. Understanding where it stands matters for any actuary whose work touches underwriting systems.

NAIC Model Bulletin and State Adoption

The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence by Insurance Companies in December 2023. The bulletin establishes guidelines and expectations for responsible AI use aligned with the NAIC's Principles of Artificial Intelligence, covering governance, risk management, transparency, and fairness. By early 2026, more than half of all states had adopted the NAIC Model Bulletin or substantially similar guidance, creating a regulatory baseline that carriers and their actuarial teams must navigate in most major markets.

The 12-State AI Evaluation Tool Pilot

The more operationally significant 2026 development is the NAIC AI Systems Evaluation Tool pilot, launched across 12 states on March 2, 2026. The participating states are Colorado, Maryland, Louisiana, Virginia, Connecticut, Pennsylvania, Wisconsin, Florida, Rhode Island, Iowa, Vermont, and California. The tool provides regulators with a structured framework for reviewing insurer AI systems during market conduct examinations.

The tool is organized around four exhibits. Exhibit A requires insurers to quantify their AI usage. Exhibit B is a governance risk assessment framework. Exhibit C covers details on high-risk AI systems (those regulators prioritize for review given potential consumer or financial impact). Exhibit D addresses AI data details. During the pilot, regulators are applying proportionality: higher scrutiny for high-risk consumer-facing systems, lighter touch for low-risk back-office applications.

The pilot runs through September 2026. The NAIC expects to update the tool based on pilot feedback in the October 2026 timeframe and bring it forward for formal adoption at the fall 2026 national meeting. This timeline means carriers in the 12 pilot states are operating under active regulatory examination of their AI systems now.

What the Regulatory Timeline Means for Actuaries

Actuaries who sign rate filings or certify underwriting models in pilot states should understand what regulators are examining under the AI Evaluation Tool. Exhibit C, which covers high-risk AI systems, is directly relevant to rating models and underwriting algorithms that affect individual risk classification. The actuary's certification that a model is actuarially justified and non-discriminatory is precisely what regulators are seeking to verify through this framework.

The current NAIC approach does not replace existing rate filing review. It supplements it with a governance and fairness assessment that the actuary cannot simply delegate to the technology or data science teams.

The Three Ways AI Actually Changes Actuarial Work

The impact on actuarial practice is not abstract. It breaks into three distinct, concrete areas, each with specific professional standards that already apply.

1. Model Validation Under ASOP 56

ASOP 56 (Modeling, effective December 2020) governs how actuaries build, review, and use models in their work. It requires documentation of model limitations, validation of model results, disclosure of significant uncertainties, and evaluation of whether the model is appropriate for its intended purpose. These requirements apply to AI and machine learning models used in actuarial work, not just traditional GLMs or frequency-severity models.

The practical challenge is that AI models, particularly gradient boosting ensembles and neural networks used in underwriting, may be less transparent than the models ASOP 56 was primarily written for. Evaluating lift, calibration, stability, and potential bias in a machine learning model requires a different toolkit than validating a chain-ladder reserve development. But the professional obligation is the same: if the actuary relies on the model output, the actuary is responsible for understanding the model's limitations and documenting them appropriately.

Patterns we have seen in exam preparation and actuarial practice over the past several years: candidates and junior actuaries frequently underestimate the documentation requirements of ASOP 56 when applied to third-party or data science-team models. The standard does not give the actuary a pass on validation because they did not build the model. "The data scientists told me it works" is not an ASOP 56-compliant model validation approach.

2. Non-Discrimination Certification Under ASOP 12

ASOP 12 (Risk Classification Systems, revised 2020) governs how actuaries develop and evaluate systems for classifying risks. The standard requires that risk classification systems be based on relevant variables, that the actuary consider potential adverse effects on protected classes, and that the actuary document the considerations applied. AI models that use large variable sets, including proxy variables that correlate with protected characteristics, create classification system review obligations that fall directly on the signing actuary.

Multiple states have enacted specific guidance on algorithmic bias in insurance, and the NAIC Model Bulletin explicitly addresses non-discrimination requirements for AI-driven underwriting. The actuary certifying a rate filing in any of the more than 25 states that have adopted the Model Bulletin or similar guidance is certifying that the AI component of the rating system meets these requirements. That certification cannot be outsourced to the technology team.

3. New Data Integration and ASOP 23

AI is generating data streams that did not exist in traditional actuarial practice: telematics data from connected vehicles, satellite and aerial imagery for property inspections, natural language processing on claims notes, and sensor data from IoT devices. ASOP 23 (Data Quality, revised 2016) governs how actuaries evaluate, use, and disclose data limitations in their work. It requires the actuary to assess the data for reasonableness, consistency, and appropriateness for the intended use.

Integrating novel data sources into pricing and reserving models creates ASOP 23 obligations. If a carrier uses satellite imagery to classify roof condition and the actuary signs the rate filing, the actuary must be able to assess whether the imagery data is of sufficient quality, collected consistently, and appropriate for the rating variable being constructed from it. The data science team's confidence in the model does not substitute for the actuary's independent data quality assessment.

The global AI model risk management market context provides some scale: the market reached approximately $6.43 billion in 2025 and is projected to reach $28.54 billion by 2035, growing at a compound annual rate of 16.2 percent, according to DataM Intelligence. The investment reflects how seriously companies in financial services are taking model governance, and insurance is a core part of that investment.

What the NAIC Survey Data Shows About Current Adoption

The NAIC's surveys of insurers by line of business provide granular baseline data on where AI adoption actually stands in 2025-2026:

Line of Business	Insurers Responding	Use, Plan to Use, or Plan to Explore AI/ML
Auto	193 insurers	88%
Health	93 insurers	92%
Home	194 insurers	70%
Life	161 companies	58%

The variation by line is meaningful. Auto and health have the highest adoption rates, reflecting the data availability, modeling maturity, and competitive intensity in those markets. Life insurance's lower rate reflects longer product time horizons, greater regulatory scrutiny, and the comparatively conservative culture around model innovation in life actuarial practice. Home insurance's 70 percent figure is being driven significantly by property imagery and catastrophe exposure modeling applications.

The key implication: AI in underwriting is already the norm in auto and health insurance. Actuaries in those lines who are not already engaged with AI model governance are behind the current practice standard, not ahead of a future trend.

Why AI Does Not Replace Actuaries

An actuarial opinion, a reserve certification, a rate filing: these require a credentialed actuary's signature and professional accountability. AI cannot sign a Statement of Actuarial Opinion. AI cannot appear before a state insurance department to defend a rate increase. AI cannot exercise the professional judgment required under ASOP 1 to determine whether a deviation from standard practice is appropriate for a specific situation.

What AI changes is where the actuary's time and judgment are consumed. Instead of spending three days pulling data, cleaning it, and formatting it for analysis, the actuary spends three hours. The time freed by automation is not just savings; it is reallocation toward higher-value work: interpreting model behavior, understanding tail risks, communicating results to regulators and business leadership, and exercising the professional judgment that cannot be automated.

The SOA's April 2025 report on AI risk management frameworks is explicit on this point: if you use generative AI tools in your actuarial work, you bear the same professional responsibility for the output as you would for any model. Documentation of how AI was used, what validation was performed, and what limitations were identified is not optional. This is not a future standard being developed. It is the current standard applied to a new tool category.

Who Should Be Concerned vs. Who Should Lean In

The actuaries with genuine reason for concern are those whose entire value proposition is manual data processing: running the same report every quarter, formatting data for another team's analysis, executing processes that require accuracy but not judgment. That category of work is being automated at large carriers, and the process is accelerating.

The actuaries who benefit from AI adoption are those whose work involves judgment, interpretation, regulatory communication, and professional accountability. Those responsibilities are not going to AI. They are being amplified by AI because the volume and complexity of models requiring actuarial oversight is increasing faster than the actuarial workforce can grow to meet it. The actuaries who lean in, developing validation skills, understanding governance frameworks, and integrating new data sources, are going to find themselves in higher demand with stronger compensation.

AI as a Knowledge Transfer Mechanism in a Talent-Constrained Market

This is where the AI story and the insurance talent crisis converge. BLS projections indicate the U.S. insurance industry will lose approximately 400,000 workers through attrition by 2026. The median age of the insurance workforce is 44, compared to 42 for the overall U.S. workforce, and only 25 percent of insurance workers are under age 35. Actuaries are among the roles most affected because senior actuarial expertise cannot be quickly replenished from adjacent fields.

Accenture's survey found that 76 percent of executives anticipate AI will streamline knowledge transfer between experienced underwriters and new hires. When a 30-year veteran retires, their institutional knowledge about evaluating complex risks does not have to leave with them if AI systems have been trained on their decision patterns and can surface comparable examples for junior staff.

For students and early-career actuaries, this dynamic is favorable. The talent shortage creates strong demand for anyone who combines quantitative skills with the ability to work with both traditional actuarial methods and modern AI tools. Carriers and consulting firms are competing on exam sponsorship, study time allowances, and compensation to attract the credentialed actuarial talent that AI governance work requires.

The Accenture survey's finding that 81 percent of executives believe AI will create new roles rather than simply eliminating existing ones is consistent with what we have seen in practice. The new roles are concentrated in model governance, AI ethics review, and the integration of emerging data sources. These are jobs that require an actuarial credential, technical fluency, and strong communication skills simultaneously. They are not roles that exist cleanly in either the historical actuary job description or the data scientist job description.

The ASOP Framework: What Actuaries Must Know

The relevant professional standards for AI work in actuarial practice are already in effect. This is not a future standards development question; it is an application question.

ASOP 56 (Modeling): Requires actuaries to understand material assumptions and limitations of models they rely on, document model validation, disclose significant model uncertainties, and evaluate appropriateness for the intended use. Applies to AI models used in actuarial work.

ASOP 23 (Data Quality): Requires actuaries to assess data for reasonableness, consistency, and appropriateness for intended use, and to disclose data limitations. Applies to novel data sources (telematics, imagery, IoT) integrated into actuarial models.

ASOP 12 (Risk Classification): Requires that risk classification systems be based on relevant variables and that the actuary consider potential adverse effects on protected classes. Applies directly to AI-driven rating systems.

The Actuarial Standards Board adopted a new ASOP 58 on enterprise risk management effective May 2025, and ASOP 41 (Actuarial Communications) is under revision. The pipeline of standards development reflects the volume of new practice areas the profession is formally addressing. Actuaries whose work involves AI-driven models should monitor ASB developments as they directly affect what documentation and disclosure standards apply to new model categories.

What Actuaries Should Be Building Now

Based on where regulatory requirements and employer demand are converging in 2026, three skill areas have consistent prioritization across surveys, job postings, and professional guidance:

Python and model evaluation proficiency.
You do not need to build models from scratch. You need to be able to read code, run models, evaluate gradient boosted model output, and assess lift, calibration, and stability. The threshold question is not "can you build a neural network" but "can you tell whether a machine learning model is performing as expected and document the limitations in a way that satisfies ASOP 56 and survives regulatory review." For most actuarial roles, that requires intermediate Python, not advanced software engineering. The premium for this combination in the current market is approximately 10 to 15 percent above peer compensation based on 2026 recruiting data.
AI governance and model risk management frameworks.
The NAIC Model Bulletin, the AI Systems Evaluation Tool pilot, and the SOA's AI risk management framework report (April 2025) provide the regulatory and professional framework. Understanding how these frameworks translate into actuarial obligations, what documentation is required, how to structure a model review for a state examination, is rapidly becoming core actuarial competence in auto, health, and P&C lines. The CAS and SOA have jointly published an AI competency framework that is worth reading as a baseline curriculum for self-development in this area.
Communication skills for technical translation.
When a state regulator asks whether the AI underwriting model is non-discriminatory, the actuary needs to translate calibration metrics, feature importance distributions, and protected class adverse impact analysis into clear, defensible regulatory language. When the CEO asks whether the AI claims model is performing as expected, someone needs to translate the technical metrics into business language. Actuaries who can do both are in the room for decisions that shape the company's direction. Actuaries who can do the technical work but cannot communicate it lose the governance role to someone who can.

The Bottom Line

AI in underwriting is not a future event. It is the current state of the market in auto and health insurance and is moving rapidly through commercial P&C and personal lines. The NAIC's 12-state evaluation tool pilot makes the regulatory framework concrete: insurers in major markets are having their AI governance examined now, and the actuary's certification of rate filings and underwriting models is at the center of what regulators are evaluating.

The professional standards framework, ASOP 56, ASOP 23, ASOP 12, already defines what actuaries are responsible for validating and certifying. The actuaries who understand these obligations and build the technical fluency to meet them are going to find themselves in higher demand with stronger compensation. The actuaries who treat AI as someone else's problem are going to find the profession's most valuable work moving past them.