ASOP No. 41 transparency. NAIC Model #672 immutable audit. ERISA §404(c) fiduciary grounding. Every assumption sourced, every methodology disclosed — or blocked before it exits the model.
A fabricated mortality table assumption or a cited SOA study that doesn't exist isn't a minor error — it misprices risk for thousands of policyholders. Every Sturna layer maps to a specific professional standard.
Every actuarial assumption, methodology, and data source must be disclosed under ASOP No. 41 (Actuarial Communications). Sturna enforces this at the generation level — output without traceable sourcing is flagged or blocked. No undisclosed assumptions exit the model.
ASOP No. 41 — Actuarial CommunicationsEvery actuarial output — assumption sets, reserve calculations, risk model outputs — is written to an HMAC-signed append-only audit log. NAIC Model #672 requires documented actuarial review. The log satisfies that requirement by construction, not retroactive export.
NAIC Model #672 — Actuarial OpinionAI models hallucinate non-existent SOA mortality tables (VBT, CSO) and fabricate assumption rates with false precision. Sturna cross-validates every mortality table reference against the actual published SOA tables. A table reference that doesn't exist is blocked before output.
SOA 2015 VBT / 2017 CSOFor pension and retirement actuarial work, AI outputs that inform fiduciary decisions must meet ERISA §404(c) prudent expert standards. Sturna traces every claim to a verified source. Ungrounded recommendations that could influence benefit plan decisions are blocked.
ERISA §404(c) — Fiduciary ReliefThree independent verification layers run on every response: (1) completeness check — assumption set fully disclosed; (2) accuracy check — methodology and data sources verified against SOA/NAIC standards; (3) stress check — MARCH adversarial gate catches fabricated table references and methodology violations.
Systematic defense in depthThe American Academy of Actuaries and Society of Actuaries publish standards of practice that govern methodology. Sturna's knowledge corpus is grounded in current ASOP, NAIC Model regulations, SOA research papers, and AAA practice guidelines — not pattern-matched generation.
AAA Code of Professional ConductFive prompts designed to bait an ungrounded model into fabricating mortality tables, NAIC regulations, or SOA studies. Left side is live GPT-4 output. Right is Sturna. API calls are real.
Three independent verification layers operating simultaneously — not sequentially. Any single gate failure blocks the output and writes an audit entry. There is no pass-if-two-of-three.
Verifies the actuarial response is complete: assumption set fully disclosed per ASOP No. 41, all data sources named, no silent extrapolations. Partial responses with undisclosed methodology gaps are flagged before output.
Validates methodology and citations against the SOA/NAIC published corpus. Mortality table references are looked up against actual VBT/CSO tables. ASOP citations are verified against published standards. Any reference that doesn't exist is blocked.
MARCH adversarial gate — a second agent with information asymmetry reviews the output independently. It checks for plausible-sounding but non-existent standards (NAIC Model #847, SOA 2022 Enhanced VBT), fabricated q-values, and ERISA fiduciary exposure from AI-assisted benefit determinations.
The next ASOP enforcement action won't cite your firm. Sturna deploys an actuarial-tuned agent pool with verified methodology grounding, NAIC Model #672 compliant audit trail, and ASOP No. 41 disclosure enforcement — active from day 1. Deposit credits your first month. No lock-in.