The NAIC's Big Data and Artificial Intelligence (H) Working Group holds an actuarial panel on AI governance trends on July 22, 2026, paired in the same hour with a progress update on its 12-state AI Systems Evaluation Tool pilot running through September 2026.

The pairing is the signal. With Fall National Meeting adoption targeted for November, carriers have roughly 90 days after pilot findings close to build exam-ready documentation, and the agenda puts that duty on the signing actuary rather than the compliance office alone.

Key Takeaways

  • 12 states are running the AI Systems Evaluation Tool pilot through September 2026, with monthly coordination calls specifically to keep carriers from receiving duplicative, inconsistently worded requests.
  • Four exhibits carry the exam: Exhibit A quantifies AI usage, Exhibit B assesses governance risk, Exhibit C drills into high-risk systems, Exhibit D captures data details, including a Version 4.0 addition on reasonable accommodations.
  • Roughly 90 days separate the close of pilot data collection from the November adoption vote. That is the window to reverse-engineer the exhibit structure against an existing AI inventory.
  • 24-plus states and the District of Columbia have adopted the Model Bulletin since December 2023. It carries no penalty clause of its own, and examiners are already measuring AI programs against it.
  • ASOP Nos. 56, 23, 41 and 12 are what the profession currently has. The Actuarial Standards Board has not signaled intent to draft a dedicated AI standard.

What the July 22 Panel Pairs Together

The meeting format is unremarkable: a one-hour Webex session at noon Eastern, open to the public, with two stated purposes on the working group's calendar. The substance is in the pairing. Regulators scheduled the governance-trends panel in the same hour as the pilot update, so actuaries hear both items connected rather than as separate agenda lines.

It is not the group's first such panel. A comparable session on March 24, 2026 combined actuarial governance trends with an operationalization update on the Model Bulletin. The recurring format brings practicing actuaries in front of regulators to describe what governance looks like inside a carrier, then measures that description against what the pilot tool is finding in the field.

The instrument doing the work is the AI Systems Evaluation Tool, piloted across California, Colorado, Connecticut, Florida, Iowa, Louisiana, Maryland, Pennsylvania, Rhode Island, Vermont, Virginia and Wisconsin from early 2026 through September. Participating states run monthly coordination calls through the pilot window to avoid sending carriers duplicative, inconsistently worded requests, which is itself a signal that regulators expect real information-gathering friction once the tool goes live everywhere.

The NAIC's own clarification defines the boundary precisely: "the tool does not create new requirements for AI governance risk assessments" (AI Systems Evaluation Tool, Version 4.0). It is a standardized examination instrument for a rule that already exists. Our walkthrough of the four-exhibit framework covers what each exhibit demands of a working pricing model.

MilestoneTimingWhat changes for actuaries
Pilot data collection (12 states)Early 2026 – September 2026Exhibits A–D tested against live carrier AI inventories
Tool revision & re-exposureSeptember – October 2026Feedback from pilot states incorporated; public comment reopens
Fall National Meeting adoption voteNovember 2026Tool becomes the standardized examination instrument nationally
July 22 actuarial panelJuly 22, 2026Regulators hear directly from practicing actuaries on governance mechanics before the vote

Where the Signature Sits

The governance expectation under the Model Bulletin is that a carrier's AI committee include actuarial, data science, underwriting, claims, compliance and legal representation, with senior management or board-level accountability for the program (Crowell & Moring, 2026). That is a committee structure. It answers who sits in the room, not who signs their name to a specific number.

Actuarial practice already has a precedent for the second question. The Appointed Actuary issuing a statutory opinion attests annually to having met the continuing education requirements under Section 3 of the U.S. Qualification Standards before that opinion can stand. That is a formal, individually signed attestation layered on top of a broader firm compliance program, and it is the closest existing analogue to an AI-specific actuarial attestation: not a replacement for the governance committee, but a personal certification sitting on top of it.

The pricing-side case makes the split concrete. When a predictive pricing model drifts between filing cycles without triggering a re-validation, the exposure does not sit with whoever approved the vendor contract. It sits with the actuary who certified the resulting rate indication as still representing the intended risk relationship. The committee can document that a model was deployed correctly. Only the actuary can certify that its current output still supports the number in the filing.

That is also how a guidance document acquires teeth. The Model Bulletin was adopted in December 2023 and taken up by 24-plus states and the District of Columbia through their own bulletin, regulation or circular letter processes. It carries no statutory penalty provision. Its documentation expectations, covering methodology, data sources, validation results, bias testing and ongoing monitoring, become examined-against criteria the moment a department writes them into exam procedures, and the Evaluation Tool makes that conversion systematic across 12 states at once.

The Standard That Does Not Exist Yet

The regulatory calendar is running ahead of the profession's own standard-setting process, and the distance between them is where a signing actuary is currently exposed.

The SOA Research Institute's Actuarial Intelligence Bulletin put the open question directly in its January 2026 edition: does the profession need a dedicated AI-specific ASOP, a near-term update to existing standards, or can it rely on current professionalism guidance as it stands? The Actuarial Standards Board has not signaled intent to draft a dedicated AI standard.

The working assumption is therefore that ASOP Nos. 56, 23, 41 and 12, covering modeling, data quality, communications and risk classification, carry AI-specific practice through the near term, supplemented by non-binding guidance rather than a new enforceable standard. None of the four was written with a vendor-supplied generative model in mind, so the mapping is a matter of interpretation rather than instruction.

The American Academy of Actuaries' Competency Framework began governing new-member requirements on January 1, 2026, covering general actuarial topics, U.S. laws and practices, and professionalism. It does not carve out AI governance as its own competency line distinct from general modeling literacy.

Put the two side by side. The professional bodies are still at the stage of asking whether a new standard is warranted, while the regulator is piloting a four-exhibit examination tool with a November adoption date. An actuary waiting for an ASOP to define adequate AI governance documentation will be asked for that documentation first, and will have to defend whatever they built against a standard that does not yet exist in writing.

Further Reading

Sources