By James W. Moore
Editor’s Note
Last week, Guidewire’s proposed acquisition of hyperexponential illustrated how seriously established insurance technology providers are taking AI-driven underwriting. This week offered a different perspective: evidence of what the technology is accomplishing in production, and a clearer view of how the industry is beginning to govern it.
Hiscox reported that an AI-augmented quoting tool improved speed to quote by approximately 75% in two specialty lines. Kalepa raised $75 million to expand technology that connects underwriting, pricing, and operational workflows. Aon and SAGE announced a crop insurance program that uses AI to distinguish risks at the individual field level.
Meanwhile, regulators and reinsurers are working through the consequences. The NAIC received another round of comments on its AI Risk Evaluation Supplement, while Singapore’s reinsurance industry introduced guidance addressing AI-related exposures and potential portfolio concentrations.
These developments reflect two related changes. AI is becoming part of how insurers perform their work, and increasingly part of how they understand the risks they assume.
Both deserve attention.
James W. Moore
Editor-in-Chief, InsuranceIndustry.ai
NAIC Comments Reveal Continuing Disagreement Over AI Examination Scope
The National Association of Insurance Commissioners has published 21 public comment submissions on version 6.0 of its proposed AI Risk Evaluation Supplement, following the October 30 comment deadline.
The submissions show that regulators still face consequential questions about how broadly the supplement should apply and what examiners should expect insurers to demonstrate.
One disagreement concerns generalized linear models (GLMs), longstanding actuarial tools used in insurance ratemaking. Version 6.0 allows regulators discretion to exclude certain ratemaking GLMs from the supplement’s more detailed examination exhibits.
The InsurTech Coalition argues that GLMs used in filed rates and risk classifications should be excluded throughout the supplement, including its initial inventory. The National Association of Mutual Insurance Companies (NAMIC) argues that ratemaking GLMs should be removed from the supplement’s scope entirely.
Other commenters focus on agentic AI. Several submissions ask for clearer distinctions between systems that recommend actions and systems authorized to execute them. They raise questions about meaningful human approval, the ability to reverse automated actions, and the records necessary to reconstruct individual decisions.
The supplement remains under development. The working group held another public discussion on November 2 and has scheduled a session for November 16 during the NAIC Fall National Meeting.
Why it matters: The debate is no longer simply about whether insurers should govern AI. It is increasingly about which systems require which kinds of examination.
Traditional actuarial models already operate within established regulatory frameworks. Applying overlapping governance processes without identifying additional risk could create work without improving oversight.
Agentic systems present a different problem. An insurer may authorize a system to perform activities that previously required human judgment or approval. Examiners then need to understand the scope of that authority, how it is constrained, and whether the insurer can reconstruct what happened.
The eventual supplement will help determine how regulators distinguish between models that inform decisions and systems that can act upon them.
Hiscox Reports 75% Improvement in Speed to Quote
Hiscox reported a substantial operating improvement from its AI-augmented submission-to-quote technology in its November 3 trading statement.
The insurer said the tool improved speed to quote by approximately 75% in terrorism and architects and engineers coverage within its U.S. broker business. Hiscox also reported that the technology contributed to an improved quote-to-bind ratio.
The company is extending the solution more broadly across its Retail operations.
The result is meaningful, although its scope matters. Hiscox reported the improvement in two specific lines, not across its entire underwriting operation. The statement does not establish that underwriting expenses declined proportionately or that the technology independently produced an equivalent increase in profitability.
Hiscox also reported broader technology initiatives, including an AI-native U.S. direct portal, new broker-facing systems, and continued development of agentic AI capabilities.
Why it matters: Speed to quote is a measurable operational result, particularly in broker-distributed specialty insurance where responsiveness can influence placement decisions.
The reported improvement also illustrates the value of targeting an identifiable workflow rather than attempting to transform an entire insurance organization simultaneously.
Processing submissions faster creates opportunities, but the ultimate measure is whether the insurer improves its ability to select, price, and retain desirable business.
A faster quote is useful. A faster, appropriately priced quote that wins profitable business is more valuable.
Singapore Reinsurers Introduce AI Risk Handbook
The Singapore Reinsurers’ Association launched a Reinsurers’ Artificial Intelligence Handbook on November 2, extending AI governance discussions into the risks reinsurers assume from their cedents.
Developed by a working group drawn from the association’s Technical and Legal & Compliance Subcommittees, the handbook offers a principles-based approach to assessing AI-related risks.
It addresses data and model reliability, third-party dependencies, underwriting, claims, and portfolio concentration. The guidance addresses questions specific to reinsurance.
Those questions extend beyond whether a reinsurer uses AI internally.
An insurer’s deployment of AI may alter its risk selection, pricing practices, claims outcomes, or operational dependencies. Reinsurers may ultimately participate in the financial consequences without controlling the underlying technology.
Common reliance on models, vendors, or data sources could also create exposures shared across otherwise separate insurance portfolios.
Why it matters: Reinsurers have traditionally evaluated cedents’ underwriting practices, portfolio composition, and risk management capabilities.
AI adds another dimension to that evaluation.
When multiple insurers rely on similar automated decision processes, a common weakness could affect more than one portfolio. The relevant exposure may not be visible simply by examining geography, industry, or traditional lines of business.
The handbook is guidance, not a new regulatory requirement. Its significance lies in recognizing that AI governance is becoming relevant to the transfer and accumulation of insurance risk, not merely an insurer’s internal technology controls.
Aon and SAGE Bring Field-Level AI to Crop Insurance
Aon and agricultural intelligence company SAGE announced an AI-supported precision crop insurance program on October 29, developed in collaboration with specialty agricultural insurance provider AgriCap.
The program is intended for corn and soybean growers across 15 states for the 2027 crop year.
Unlike traditional approaches that rely heavily on county-level agricultural risk assessments, the program uses data concerning soil, water, weather, and farming conditions to differentiate risk at the individual field level.
According to the companies, the technology combines datasets that historically have been difficult to analyze together at scale.
The initiative builds on a four-county Michigan pilot involving more than one million agricultural acres, developed by the RALIC coalition with state support and involving Aon and SAGE. The resulting private insurance coverage is designed to complement federal crop insurance rather than replace it.
The announced broader program remains prospective, and its ultimate underwriting performance has not yet been demonstrated.
Why it matters: Much of the insurance industry’s AI investment has focused on doing existing work more efficiently.
This initiative illustrates a different opportunity: changing the level of detail at which a risk can be evaluated and priced.
County-level averages can conceal meaningful differences between individual farms. More granular information could allow insurers to recognize risk characteristics and mitigation efforts that broader classifications overlook.
The challenge will be determining whether those distinctions consistently predict insured losses and support economically sustainable pricing.
AI does not eliminate underwriting uncertainty. Its potential here is to make distinctions that conventional data and analytical processes could not practically support.
Kalepa Raises $75 Million to Expand Insurance AI Platform
Insurance technology company Kalepa announced a $75 million Series B financing round on November 5 at a valuation of $470 million, according to Kalepa.
The financing was led by Updata Partners, with participation from existing investors IA Ventures and Inspired Capital. The company has now raised $91 million in total.
Kalepa develops AI technology intended to coordinate work across insurance operations, underwriting, pricing, and portfolio management. Its customers include W.R. Berkley, Bowhead Specialty, and AmRisc.
The company reported that work processed through its platform quadrupled over the past year. It also cited customer results including 30% to 60% more premium per underwriter, a 58% reduction in quote time, and improvements in combined ratios.
Those performance figures are company-reported and should not be interpreted as independently established results across Kalepa’s customer base.
Kalepa plans to use the financing to expand its product suite, enter additional markets, and increase staffing.
Why it matters: The financing is another indication that investors see value in insurance-specific AI platforms that connect multiple operational functions.
The larger competitive question concerns where those capabilities will reside.
Established insurance platforms already manage substantial portions of the underwriting and policy lifecycle. Specialist AI providers are developing technologies that can coordinate activities across those systems.
As AI becomes more deeply integrated into insurance operations, insurers will have to evaluate the economics of specialized platforms against capabilities increasingly incorporated into their core technology.
The investment establishes market interest. The more important test will be whether these systems produce durable improvements in underwriting quality and operating performance.
OCC Introduces Agentic AI Examination Framework for Banks
The Office of the Comptroller of the Currency issued a supervisory framework on November 4 addressing banks’ use of what the agency calls “Super Intelligence,” including agentic systems. The terminology reflects a September executive order directing federal agencies to use that term.
Although the framework applies to bank supervision rather than insurance regulation, its approach offers a useful comparison for insurers developing their own AI governance practices.
The OCC directs examiners to evaluate AI within established risk management structures rather than automatically treating it as a separate category of risk.
The framework addresses permitted uses, risk classification, human oversight, governance accountability, testing, third-party relationships, operational resilience, and documentation.
The framework directs examiners to classify risk according to the consequences of an activity, the degree of agency transferred to AI, and the extent of human oversight, rather than technical sophistication alone.
The OCC expects the framework to evolve as capabilities and supervisory experience develop.
Why it matters: The framework reflects an approach to AI governance that starts with the activity being performed and the consequences of failure.
For insurers, a similar distinction arises between an AI system that prepares an underwriting recommendation and one authorized to alter a policy, issue a payment, or execute another consequential transaction.
The relevant controls depend partly on what the system is permitted to do, how its actions are monitored, and how effectively the organization can intervene.
The OCC’s framework creates no new supervisory expectations for banks, nor does it establish insurance regulatory requirements. It does demonstrate how another financial regulator is attempting to apply existing supervisory principles to increasingly autonomous technology.
Independent Assessment Finds Acceleration in Anthropic’s AI Development
Independent AI evaluation organization METR published an assessment on November 5 examining how AI is affecting the pace of research and development at Anthropic.
Following an evaluation conducted between August 12 and October 29, METR estimated that Anthropic’s progress on well-specified model capabilities was occurring at approximately 1.9 times its earlier rate.
The organization attributed much of that acceleration to Anthropic’s use of AI in its own research, while emphasizing considerable uncertainty about the estimate and its causes.
The assessment also contains an important qualification.
Anthropic provided METR with extensive access to internal information but exercised rights to redact substantial portions of the evidence underlying some conclusions. METR disclosed those limitations and acknowledged that outside readers could not independently evaluate all the supporting material.
METR also reported seeing no clear bottlenecks that would prevent further acceleration over the next 12 months, while emphasizing that it was not confident in that assessment. The research does not establish that AI development will continue accelerating at the same rate.
Why it matters: Insurers are increasingly being asked to assess risks associated with technologies whose capabilities and operating characteristics may change rapidly.
Independent evaluations can provide information beyond developers’ own statements, but access and disclosure restrictions affect how much confidence outside parties can place in the findings.
For insurers considering AI-related liability, technology dependence, or long-term exposures, that distinction matters.
Independent assessment is valuable. Understanding what was evaluated, what remains uncertain, and which evidence can be examined is part of determining how much reliance the assessment deserves.
Sources
- NAIC, AI Risk Evaluation Supplement v6.0: Public Comments Received, November 2026.
- NAIC, Big Data and Artificial Intelligence (H) Working Group, November 2026.
- Hiscox, Q3 2026 Trading Statement, November 3, 2026.
- Singapore Reinsurers’ Association, AI Handbook, November 2, 2026.
- Aon, Field-Level Agricultural Intelligence for Crop Insurance, October 29, 2026.
- Kalepa, $75 Million Series B Announcement, November 5, 2026.
- OCC, Supervisory Framework on Bank Use of Super Intelligence, November 4, 2026.
- METR, Independent Assessment of AI R&D Acceleration at Anthropic, November 5, 2026.
More AI Insights
Catch up on previous editions of AI Insights, our weekly review of the developments shaping AI and insurance.
Browse AI Insights Archive
AI Disclaimer: This content was created with assistance from artificial intelligence technology. While content is based on factual information from the source material, readers should verify all details directly with the respective sources before making business decisions.
