Gen Re says AI underwriting must balance speed, accuracy and trust

Gen Re issued AI-Enabled Underwriting 2.0 - Balancing Efficiency, Accuracy, and Trust on August 4, 2026, as carriers push for faster decisioning and less manual work without giving up accuracy, governance, or trust. That tension is sharpest in commercial and specialty lines, where one weak decision can echo across a portfolio.

Underwriting 2.0 starts with workflow, not model demos

The useful version of underwriting AI is not a chatbot bolted onto a policy admin screen. It is the layer that surfaces relevant data, prioritizes submissions, summarizes exposures, identifies missing information, and supports referral logic inside the underwriter’s daily workflow. In Gen Re’s framework, AI is decision support, not a replacement for the people who have to stand behind the decision.

That distinction matters when software is judged in a live sandbox or a real production team. If the system adds another interface, another queue, or another set of unexplained scores, it creates more noise than value. The stronger platforms are the ones that turn unstructured submission data into a clear next action, then preserve the underwriter’s ability to override, escalate, and document why.

Regulation already sits inside the product decision

The compliance backdrop is no longer abstract. The National Association of Insurance Commissioners updated its Artificial Intelligence background page on April 3, 2026. AI is used across many industries, including insurance. The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers on December 4, 2023, and by March 2025 the NAIC counted 24 states that had adopted it with little to no material changes.

Massachusetts moved on this early as well. The Massachusetts Division of Insurance issued Bulletin 2024-10 on December 9, 2024, specifically on the use of artificial intelligence systems in insurance. That regulatory line tells carriers exactly how underwriting AI will be judged: not just on whether it works, but on whether the carrier can explain, monitor, and govern it.

For software buyers, this changes the spec. An underwriting tool is no longer just a productivity purchase. It has to support documentation, control points, and escalation paths that satisfy internal governance, state oversight, and downstream scrutiny from distribution partners and policyholders.

Explainability has to live in the workbench

AI in underwriting only scales if the outputs are explainable enough for internal governance and credible enough for regulators. A 2022 academic article on explainable artificial intelligence in insurance, published in MDPI’s Risk journal, focused on the same pressure: opaque decisions do not survive contact with real oversight. That is especially true when an underwriter has to justify a declination, a surcharge, or a referral.

The control problem is bigger than model choice. Carriers need data quality checks, model monitoring, human oversight, and a clean escalation path when confidence is low. The software has to make those controls operational, not decorative. If the workbench does not show why a model pushed a submission into referral, when an underwriter should override it, and how the exception gets reviewed, then the AI is making decisions that the business cannot defend.

The adoption curve is already past the experiment stage

A 2022 NAIC survey showed that 88% of private passenger auto insurers and 70% of homeowners insurers already use or plan to use AI or ML models in underwriting; Quick Silver Systems cited the survey in August 2025.

An April 2026 article by IHS and Sam Houston State University described AI risk in insurance as dynamic, opaque, and only partially understood. That language fits the underwriting use case perfectly: the risk environment changes, the model logic is not always obvious, and partial understanding is not enough when a carrier has to stand behind a file on the record.

Wilson Elser wrote on January 6, 2026, that insurers are pushing hard to invest in AI systems that improve growth and operational efficiency, while governance teams scrutinize how those systems make decisions. That split is the core management problem in underwriting 2.0: the business wants lift, but the control function wants proof.

What carriers should pressure-test in vendor demos

The right underwriting platform should be judged on how it handles trust controls inside the workflow, not on how glossy the demo looks. The test is whether the system can support a faster decision without forcing the underwriter to guess at the logic or burying exceptions in a separate compliance queue.

Look closely at whether the platform can do these things:

• Surface the submission data that matters, then explain why it matters
• Prioritize risks with clear referral thresholds rather than opaque scores
• Flag missing information and route it back into the underwriting flow
• Preserve underwriter overrides, with reasons attached to the file
• Log model changes, monitoring outputs, and exception handling for governance review
• Handle low-confidence cases by escalating them instead of forcing a false precision

That is where many products break. They can summarize a loss run or rank a submission, but they cannot show the logic chain in a way that satisfies a commercial lines underwriter, a compliance lead, and a regulator at the same time.