The SOA and CAS have both moved AI from discussion to curriculum inside eighteen months. The SOA launched its Actuarial Intelligence Bulletin in March 2025 and published six editions through January 2026, building a competence framework around the existing exam pathway. The CAS built a separate eight-session bootcamp for actuaries who credentialed before machine learning appeared on any syllabus. The two efforts solve different halves of the same problem, and only one of them scales.

Key Takeaways

  • Eight sessions over eight weeks in the CAS AI Fast Track, delivered with Akur8, capped at 200 members per cohort with at least three cohorts run through 2025.
  • $550 for CAS and iCAS members, $750 otherwise, earning a Certificate in Advanced AI for Actuarial Science and up to 9 CE credits.
  • A 96-hour take-home assessment at $1,255 is how the SOA tests applied machine learning in Advanced Topics in Predictive Analytics, required for the FSA in several tracks.
  • Under 1% actuarial unemployment against 22% projected job growth from 2023 to 2033, which is the market condition the credentialing changes are responding to.
  • Fewer than 50% of practicing actuaries demonstrate data science and AI proficiency while over 60% name it as a critical gap.

What the Two Bodies Actually Shipped

The SOA's framework arrived through its bulletin series rather than a single document. The September 2025 edition covered trustworthy models, SHAP explainability, LLM use, reserving automation, and introduced a skill development competency framework. The January 2026 edition by Carlos Arocha, FSA, set it out most explicitly, organized around three forces: data volume, computational power, and stakeholder expectations on AI governance.

The exam pathway carries the lower rungs. Exam SRM covers regression, time series, principal components, decision trees, and clustering. Exam PA adds GLMs, tree-based models, and unsupervised learning with a hands-on component. Advanced Topics in Predictive Analytics sits at the FSA level covering gradient boosting, neural networks, and advanced unsupervised learning, assessed through a 96-hour take-home format at $1,255 that requires building and validating models rather than describing them.

The CAS took the other route. Its AI Fast Track Program, run with Akur8's actuarial data science team, is a virtual bootcamp of eight sessions over eight weeks, capped at 200 members per cohort, with at least three cohorts completed through 2025. Participants earn a Certificate in Advanced AI for Actuarial Science and up to 9 CE credits at $550 for members and $750 for non-members.

The session sequence is deliberate. It opens by framing AI as a collection of algorithms rather than magic, moves through retrieval-augmented generation, rules-based systems and reinforcement learning, then reaches its technical core in a machine learning session on why modern GLM variants and gradient boosting machines suit insurance problems. Deep learning follows, explicitly teaching participants to look inside neural networks rather than accept the black-box framing. Generative AI covers hallucination mitigation, and ethics closes the program.

On the CAS exam side, MAS-I and MAS-II carry extended linear models and credibility, the Property-Casualty Predictive Analytics component sits in the ACAS pathway, and the CAS Institute's CSPA credential offers a five-course specialization.

The Hiring Market Set the Baseline First

DW Simpson's 2025 Market Trends report describes Python, R, and SQL as baseline expectations rather than differentiators. Its September 2025 analysis put actuarial unemployment under 1% with 22% projected job growth from 2023 to 2033, roughly five times the national average, and framed the role shift from behind-the-scenes analyst to strategic leader as something the technical skills enable rather than accompany. The salary premium attached to hybrid data science roles is the same signal priced.

The deployment data explains where the demand comes from. A 2025 NAIC survey of 93 health insurance companies found 84% using AI or machine learning, with 92% holding governance principles aligned to the NAIC AI Principles. A Novarica survey of 51 North American CIOs put 59% with ML implemented in actuarial processes. Goldman Sachs reported 42% of American insurers using AI by 2023. Every one of those deployments needs someone accountable for validating it.

The operational consequence is a change in what the job consists of. Research from hyperexponential puts the pre-AI split at roughly 70% of an actuary's time on data manipulation and calculation against 30% on interpretation, inverting toward 30% on validation and monitoring and 70% on interpretation and stakeholder communication.

That inversion is what the curricula are actually pricing. Technical excellence in a spreadsheet workflow meant building the model faster. In an AI-augmented workflow it means recognizing that a model's output is wrong, locating why, and defending that judgment to a regulator. A carrier running a proprietary agentic platform or a build-versus-buy LLM stack needs actuaries who can interrogate it technically, not describe it conceptually. The convergence is international: the SOA and the UK's Institute and Faculty of Actuaries have both updated syllabi toward data science and predictive analytics, and the IAA published its governance framework, model testing paper, and cross-country regulatory comparison in November 2025.

The Rungs Do Not Reach the People Making the Decisions

The exam changes work on candidates. A candidate credentialing in 2026 has demonstrated machine learning proficiency that was not tested five years ago. But the actuaries deciding whether a gradient boosting model may support a rate filing are mostly not candidates. They passed their last exam before SHAP values appeared in any actuarial curriculum, and no exam requirement reaches them.

The size of that gap is documented. Fewer than 50% of practicing actuaries demonstrate data science and AI proficiency, while over 60% name it as a critical skill gap. That is an awareness gap closing faster than a capability gap, which is the harder configuration: senior practitioners who correctly identify what they cannot do are still the ones signing off.

The remedies are throughput-limited by design. The CAS Fast Track runs at 200 seats a cohort against a profession several orders larger, and at least three cohorts had completed through 2025. It is a well-built program at a low price point; it is not a mechanism for retraining a generation. The SOA's route runs through PD Edge+ and CPD requirements steered toward AI governance, which reaches everyone but is self-attested rather than assessed.

Former Academy president Darrell Knapp put the professional consequence directly, describing existing actuarial standards as a reasonable set for AI use while cautioning that shortcuts become dangerously amplified with automation. That is the exposure the throughput problem creates. An actuary relying on a model must understand its limitations and document that reliance, and the number of seats available to build that understanding is fixed while the number of deployed models is not.

The BLS projects 22% actuarial employment growth through 2034 with roughly 2,400 annual openings, and the Academy's own framing names competition from non-credentialed practitioners performing analytical work actuaries historically did. The credential's defense is the accountability layer rather than the technique, and that defense holds only where the person carrying it can evaluate what they are accountable for.

Further Reading on actuary.info