Large Language Models (LLMs) and the Actuarial Craft

Advertise - Write For Us

Large Language Models are reshaping actuarial work, not by replacing judgment, but by removing friction. Here’s how disciplined use of LLMs can boost productivity while strengthening governance, professionalism, and trust.

LLMs Hold Promise for the Actuarial Field

Large Language Models (LLMs) have emerged as one of the most transformative technologies of the decade, demonstrating remarkable capabilities in understanding natural language, generating human-quality text, analyzing complex documents, and even assisting with coding and mathematical reasoning. While much attention has focused on their applications in customer service, content creation, and software development, LLMs hold particular promise for the actuarial profession a field built on processing vast amounts of information, communicating technical concepts, and making sense of complex regulatory and business environments.

In March 2025, the Society of Actuaries (SOA) Research Institute convened a panel of experts to discuss the use of generative AI in the insurance industry. The panel concluded that current AI tools, such as LLMs, can boost productivity for some tasks, but the technology hasn’t evolved enough to replicate actuarial analysis and decision-making. However, the panel predicted it will become necessary for actuaries to use these tools .

From a practitioner’s perspective, LLMs are best understood as infrastructure rather than magic. They sit alongside spreadsheets, valuation systems, statistical software, and documentation templates. Used well, they reduce friction in work that actuaries already do. Used poorly, they introduce new operational and ethical risks. Three frameworks help ground this reality. UNESCO’s Recommendation on the Ethics of Artificial Intelligence sets out global principles for responsible AI. The National Association of Insurance Commissioners has articulated supervisory expectations specific to insurance and financial services. SOA’s report “Operationalizing LLMs: A Guide for Actuaries” translates these ideas into practical controls and workflows. Read together, they form a pragmatic lens for adoption rather than a marketing narrative.

Efficiency Without Abdication of Judgment

One of the clearest benefits of LLMs is information synthesis. Actuaries regularly work across accounting standards, regulations, guidance notes, and academic research. An LLM can ingest long documents and surface relevant sections in minutes. In an IFRS 17 context, for example, an actuary can rapidly map disclosure requirements or identify where interpretations differ across jurisdictions. This does not replace reading or interpretation. It changes where effort is spent. Time shifts from searching to thinking.

Communication is another area where LLMs already deliver value. Translating technical results for boards, audit committees, and regulators is an acquired skill. LLMs help structure narratives, improve clarity, and adjust tone. The actuary remains the author of the substance. The model becomes an editor and translator. This division of labor preserves professional responsibility while improving the quality of communication, a point often underestimated in technical professions.

The NAIC principles reinforce this approach by stressing that AI systems used by insurers must be governed, documented, and auditable. When actuaries use LLMs internally, the same discipline applies. Management should know where AI tools are used. There should be clear policies on acceptable use, data handling, and validation. From a supervisory perspective, undocumented reliance on generative tools creates risk, not innovation.

Newsletter continues after job posts…

Coding, Models, and the New Actuarial Assistant

Modern actuarial practice is inseparable from code. R, Python, SQL, and proprietary platforms are now core tools. LLMs function well as coding assistants. They help draft functions, explain unfamiliar libraries, and debug errors.

The SOA’s Guide on Operationalizing LLMs for Actuaries emphasizes a critical point here that code generated by an LLM is not validated code. It is a starting point. Actuaries must review logic, test outputs, and ensure numerical integrity. This mirrors existing practice with spreadsheets and legacy models. The difference is speed. Errors can be produced faster, which raises the importance of controls rather than reducing it.

There is also a democratizing effect. Actuaries earlier in their careers can experiment with automation and analytics more confidently. Senior actuaries can focus on model design and interpretation rather than syntax. Over time, this may reshape team structures, but it does not eliminate the need for deep technical understanding. It simply reallocates effort.

Governance, Risk, and Regulatory Expectations

The NAIC principles bring governance to the foreground. They emphasize accountability, data integrity, transparency, and risk management in AI systems. For actuaries, this translates into practical questions. What data is being shared with an LLM? Is it confidential or personal? Is the model external or internal? Are outputs stored, logged, or reused?

UNESCO’s guidance on data protection and privacy is particularly relevant. Many public LLMs retain prompts for training or monitoring. Feeding sensitive policyholder data, financial projections, or proprietary assumptions into such systems can breach confidentiality obligations. Practitioners must understand the deployment context. Enterprise tools with contractual safeguards are very different from consumer interfaces.

Another risk is overconfidence. LLMs are persuasive by design. They produce fluent answers even when wrong in their ‘hallucinations’. In actuarial work, plausible error is more dangerous than obvious failure. The operational response is the review culture. Outputs must be challenged like junior analyst work. Cross checks, reasonableness tests, and independent review remain essential.

Documentation also matters. If LLMs are used in preparing regulatory filings or valuation reports, their role should be documented internally. This does not mean lengthy disclosures for every spelling correction. It means clarity when AI materially influences analysis, structure, or interpretation. This aligns with both NAIC expectations and actuarial standards of practice.

From Principles to Practice in Actuarial Work

UNESCO’s AI Recommendation emphasizes human oversight, accountability, transparency, fairness, and robustness. These ideas align closely with actuarial professionalism. Actuaries are already trained to document assumptions, justify methods, and stand behind results. Human oversight means more than reviewing outputs. It means understanding where LLMs are used in a workflow, what risks they introduce, and where judgment must intervene.

In practice, this means LLMs are most appropriate in supporting roles. Summarizing regulatory texts, drafting first versions of reports, restructuring documentation, or assisting with code scaffolding are low risk entry points. These uses respect the principle that final decisions remain human. UNESCO’s emphasis on traceability and explainability also resonates. If an LLM materially influences an actuarial output, that influence should be knowable and defensible. Black box reliance is incompatible with actuarial accountability.

The actuarial reality is that time pressure is constant. Deadlines, regulatory submissions, and commercial expectations push professionals toward efficiency. LLMs offer relief here, but only if their use is disciplined. Ethical AI in actuarial practice is about building habits that preserve review, challenge, and professional skepticism.

Looking Forward

LLMs represent early stages of a broader AI transformation affecting virtually all knowledge work. The specific capabilities of current models will evolve, new applications will emerge, and the technology will become more sophisticated. Actuaries who develop comfort and competence with LLMs now will be better positioned to leverage future developments.

Practitioners who engage early gain an advantage. They learn where LLMs help and where they fail. They build internal guidelines that reflect real workflows rather than theoretical fears. They also shape organizational culture, ensuring AI adoption strengthens rather than dilutes professionalism.

UNESCO’s ethical framework reminds us that technology should enhance human capability and dignity. The NAIC principles remind us that insurance is a regulated trust business. The actuarial guide reminds us that tools only work when embedded in sound processes. Together, they point to a future where LLMs are ordinary, governed, and useful.

For actuaries, this is familiar territory. The profession has always integrated new tools, from mortality tables to stochastic simulations. Each innovation changed methods but not purpose. LLMs are another chapter in that history. Used thoughtfully, they allow actuaries to spend less time wrestling with text and code, and more time exercising the judgment that the public, regulators, and markets ultimately rely on.

Pulse For The Week