CAS Preliminary Exam Review Signals the Biggest Content Overhaul in a Decade

The CAS is reviewing what its preliminary exams test. The review is driven by the CAS Actuarial Professional Analysis, the practice-focused successor to the 2020 Job Task Analysis, and it follows the Admissions Transformation Plan that rebuilt the fellowship exams between 2022 and 2026.

The organizing principle is stated plainly by the CAS: "JTA results are the guide for what is tested; if it's not used in practice now or in the future, it doesn't make it into the exam." The preliminary exams are the last part of the pathway that has not been put through it.

Key Takeaways

• Statistical Learning went from 10-20% of MAS-II to 40-50% in the Fall 2023 reweighting, while Bayesian Analysis and MCMC were removed outright as "not a standalone domain required for CAS credentials."
• The PCPA became mandatory for new ACAS candidates on January 1, 2026: a 40-question, two-hour exam plus a two-week project of roughly 20 hours, at a $700 project fee per attempt.
• The PCPA content outline, dated May 2024, stops at classical supervised learning. It covers GLMs, decision trees, random forests and gradient boosting, and does not mention large language models, generative AI, causal inference or cloud-scale computing.
• 12 to 18 months is the CAS's consistent gap between announcing a content change and first administering it, so anything announced in late 2026 lands no earlier than Fall 2027.
• MAS-I pass rates have run 42.3% to 55.3% per sitting and MAS-II 49.9% to 65.0%, and first sittings of restructured exams sit at the low end while materials and question calibration catch up.

What the Last Reweighting Already Did

The clearest evidence of what this review will do is what the same methodology did to MAS-II three years ago.

The 2020 JTA found practicing P&C actuaries relying heavily on generalized linear models, regularized regression, decision trees, random forests and gradient boosting for pricing, reserving and underwriting, against a preliminary syllabus that underweighted all of it. The Fall 2023 domain weights were the response.

MAS-II Domain	Pre-2023 Weight	Current Weight
Statistical Learning	10-20%	40-50%
Credibility	5-15%	15-25%
Bayesian Analysis/MCMC	10-20%	Removed
Time Series	Not included	10-20%

The removal of Bayesian Analysis and MCMC is the more instructive half. The JTA finding was narrow and defensible: the topic was "not a standalone domain required for CAS credentials" in current P&C practice. Bayesian methods matter in some actuarial applications, but the survey data did not show them used broadly enough across P&C roles to hold a required-exam slot. Statistical Learning took the space, moving from a minor component to nearly half the exam.

The same swap has precedent above the preliminary level. When the SOA discontinued Exam MFE after Spring 2018, the CAS reassessed whether financial economics belonged in the P&C pathway at all, concluded it did not, and replaced the requirement with an online Data Concepts course covering data visualization, data quality, and structured versus unstructured data. Traditional mathematical content out, applied data skills in.

Where the Pathway Currently Stops

The PCPA is the best available statement of what the CAS considers essential applied skill, and its boundary is as informative as its content.

Mandatory for new ACAS candidates since January 1, 2026, it runs as a 40-question, two-hour multiple-choice exam on data exploration, GLM construction, decision trees and model validation, plus a two-week hands-on project of roughly 20 hours on historical P&C datasets at a $700 project fee. It tests data cleaning, feature engineering, model selection, validation and communication of results to non-technical stakeholders, which is the competency set the JTA identified as most critical.

Its content outline, dated May 2024, covers supervised learning and nothing beyond it. Large language models, generative AI, causal inference and cloud-scale computing do not appear. That boundary is the practical answer to what a CAS credential currently certifies about a candidate's AI literacy, and it is the gap the preliminary review has to decide whether to close.

The CAS is looking at it from outside the exam system. The CAS AI Working Group published its AI Primer in March 2026, and there is an open Request for Proposals on adapting large language models for specialized P&C actuarial reasoning carrying up to $40,000 in research funding. Neither touches exam content.

The SOA is running the same problem with a different answer. Its Exam PA is conceptually identical to the PCPA, a computer-based assessment of practical modeling, and Advanced Topics in Predictive Analytics pushes data science deeper into the ASA pathway, where the CAS has no equivalent. At fellowship level the SOA went the other way, launching a flexible pathway in Fall 2025 with modular technical courses across seven practice areas and grading cut from 11 weeks to four, while the CAS deepened and accelerated the exams it already had. The 2026 SOA job analysis survey is likely to reach the same conclusion the CAS analysis already has.

The Lag Is Structural, Not Administrative

The constraint on closing that gap is that every part of the machinery is calibrated to a multi-year cycle, and the input data ages while it runs.

The CAS has held to 12 to 18 months between announcement and first administration across three restructurings. MAS-I was announced in January 2017 and first sat in Spring 2018. The 2011 revision ran transition periods through May 2011 and May 2012. The ATP fellowship changes rolled out over four sittings from October 2023 through Fall 2024, with roughly 12 months of notice before each content outline took effect. Study providers then need a further six to nine months after an outline publishes to produce materials.

Stacked, that puts changes announced in late 2026 or early 2027 into candidates' hands around Fall 2027 or Spring 2028, working from a job analysis whose predecessor was fielded in 2020. Eight years from the survey that identifies a gap to the sitting that tests it is the actual cycle time, and it is why the PCPA outline stops where the tooling of 2024 stopped.

The first sittings carry the cost. MAS-I has ranged from 42.3% to 55.3% per sitting, with 5,843 passes across 12,886 attempts through August 2025, a 45.3% cumulative rate against a unique-candidate rate nearer 58.6%. MAS-II has ranged from 49.9% to 65.0%. Candidates who sat the first MAS-I in Spring 2018 were writing an exam whose difficulty had not yet been tuned against a live population.

The other constraint is shared. Exam 1/P and Exam 2/FM are jointly administered with the SOA, so the CAS cannot rewrite the front of its pathway alone, and diverging would cost the mutual-recognition economics that the Board's revised Exam Waiver Policy exists to manage. The NAIC model bulletin on insurer AI use, adopted by roughly 24 states, is meanwhile setting employer expectations on a regulatory clock rather than an admissions one.