By James W. Moore
Editor’s Note
AI is speeding up more than the work organizations want it to do. It is also accelerating the threats they must defend against and testing how effectively existing oversight and response systems can keep up.
The NAIC has adopted its AI Risk Evaluation Supplement after months of drafting, pilot testing, and public comments. UK financial regulators have published new guidance addressing the cyber risks associated with AI, including the growing difference between how quickly vulnerabilities can be discovered and how quickly organizations can correct them.
Google’s latest threat intelligence documents increasingly autonomous offensive tools. Marsh reports a sharp increase in cyber incidents among its clients, while the National Insurance Crime Bureau describes the growing use of AI in questionable insurance claims.
There is a more encouraging development as well. Marsh’s digital placement technology demonstrates how redesigning established workflows can dramatically reduce the time required to complete insurance transactions.
AI is changing how quickly organizations can operate, discover weaknesses, or manufacture convincing evidence. The systems responsible for control, verification, and response will have to adapt as well.
James W. Moore
Editor-in-Chief, InsuranceIndustry.ai
NAIC Adopts AI Risk Evaluation Supplement Following Fall National Meeting
The National Association of Insurance Commissioners has adopted its AI Risk Evaluation Supplement, completing a development process that included multiple revisions, a pilot involving 12 states, and extensive industry comment.
The Big Data and Artificial Intelligence (H) Working Group adopted the supplement on November 16 after agreeing to two technical edits raised during its meeting. The parent Innovation, Cybersecurity, and Technology (H) Committee adopted the supplement on November 17.
The adopted framework builds on the approach discussed in last week’s AI Insights. It is designed to help state regulators evaluate insurers’ use of AI systems, beginning with an inventory of relevant applications and expanding into governance, individual models, and supporting data when the identified risks warrant further examination.
One important provision concerns traditional predictive models.
Version 7.0 permits regulators to exclude some or all generalized linear models and other predictive models from the supplement when appropriate, recognizing the existing regulatory oversight provided through insurance ratemaking. The exclusion is discretionary, not automatic.
The supplement also examines the effectiveness of human oversight, including whether reviewers can reverse, modify, or escalate AI-assisted decisions, whether they have appropriate qualifications, and whether documented overrides actually occur.
The NAIC’s meeting summaries confirm adoption but do not detail the two technical edits agreed at the Working Group meeting. The November 9 version 7.0 document remains the available copy prepared for adoption. The supplement is intended as a supervisory examination resource, not a new insurance law that becomes automatically binding in every state.
Why This Matters: The NAIC has moved from developing an examination framework to making one available for regulatory use.
State regulators will retain discretion over whether and how to employ the supplement. Nevertheless, insurers now have a clearer picture of the information and evidence regulators may request when examining AI governance.
The Working Group has also outlined regulator training on the Model Bulletin and supplement, expected to conclude in early 2027. State implementation and examination practices will determine how consistently the framework is used.
UK Financial Regulators Address the Growing Gap Between Cyberattack and Response Speeds
On November 19, the Bank of England, Prudential Regulation Authority, and Financial Conduct Authority published proposed cyber-resilience guidance alongside a separate document addressing effective practices for managing AI-related cyber risks.
The regulators identify a significant change in the cybersecurity environment.
AI is reducing the time, expense, and technical expertise required to discover vulnerabilities and identify potential attack paths. Techniques that once required considerable manual effort can increasingly be accelerated through automated analysis.
The resulting challenge is not limited to detecting vulnerabilities.
Organizations must still determine which findings are important, validate the risks, develop and test corrections, coordinate changes with technology providers, deploy fixes, and verify that the vulnerabilities have been resolved.
The regulators argue that traditional approaches built around periodic vulnerability identification may no longer be sufficient. Their AI-specific publication encourages a shift toward continuous exposure management and faster execution of remediation.
It also emphasizes the importance of identifying technology dependencies, reducing unnecessary access privileges, and ensuring that boards receive evidence of whether cybersecurity controls actually work.
The announcement follows a substantial increase in reported cyber incidents. The FCA said October 2026 had the second-highest monthly number of reported cyber incidents in five years, and the highest in the previous 18 months, 52% above the prior high in May 2025.
The general cyber-resilience guidance is open for consultation until February 26, 2027. The accompanying AI-specific effective practices are separate from the consultation and are not part of the regulators’ formal expectations.
Why This Matters: Cybersecurity effectiveness increasingly depends on the time required to move from identifying a vulnerability to eliminating the resulting exposure.
For insurers, this has two implications.
First, their own technology environments face the same challenges as those of other financial institutions.
Second, the resilience of insured organizations may become increasingly dependent on their ability to remediate vulnerabilities and maintain operations, rather than simply demonstrate that security tools and monitoring programs exist.
For insurers assessing cyber resilience, the interval between identifying a weakness and actually closing it deserves greater attention.
Google Finds AI Agents Moving Into Active Cyberattacks
Google Threat Intelligence Group published research on November 16, drawing partly on Mandiant incident-response findings, documenting the expanding use of agentic AI tools in malicious cyber operations.
The report describes threat actors moving beyond using AI to generate content or assist with isolated technical tasks.
Researchers identified offensive frameworks designed to coordinate multiple activities, including reconnaissance, vulnerability analysis, exploitation, and subsequent actions within compromised environments.
The report also describes tools intended to automate analysis of software patches. Such systems can compare updated software against earlier versions to identify vulnerabilities that may remain exploitable in organizations that have not yet installed the corrections.
This approach is particularly significant because publicly released security patches can reveal information about the weaknesses they are intended to repair.
Google also documents attacks involving stolen AI-service credentials and unauthorized use of cloud computing resources. Mandiant reported single-day fraudulent AI-service charges exceeding $100,000 in some incidents and cumulative resource abuse reaching $1 million before remediation in the most severe cases.
GTIG identified an offensive framework, exploit_agent, designed to automate vulnerability discovery and exploitation without a human in the loop, including comparing patches and generating exploits. Other observed tools automate only parts of an attack; the report does not claim reliably successful autonomous intrusions across all targets.
Why This Matters: AI is beginning to change the operating economics of cybercrime.
Attackers may be able to examine more targets, test more possibilities, and pursue opportunities with fewer human resources.
For cyber insurers, this raises questions about the reliability of historical assumptions concerning attack frequency, the effectiveness of established controls, and the time organizations have available to respond.
It also creates an opportunity for defenders to use similar technologies to accelerate detection, analysis, and remediation.
Defensive uses of AI may improve response, but capability alone does not establish which side gains more.
Marsh Client Cyber Incidents More Than Double as AI Tools Spread
Cyber incident data from Marsh provides another perspective on the changing threat environment.
According to Reuters reporting published November 17, Marsh clients reported 105 cyber incidents during the final two weeks of October, compared with 48 incidents during the final two weeks of May.
That represents more than a doubling between the two reporting periods.
Meredith Schnur, Marsh Risk’s U.S. and Canada cyber practice leader, attributed a significant portion of the increased activity to attackers using AI-enabled tools to identify and exploit vulnerabilities.
The reported weaknesses included unpatched software, exposed systems, and inadequate identity controls. These are not fundamentally new vulnerabilities, but AI can make them easier to discover and exploit.
Marsh also identified the growing difficulty organizations face in containing attacks and responding to incidents.
The figures represent incidents reported by Marsh clients during two specific periods. They should not be interpreted as an industrywide measure of cyberattack growth or proof that AI caused the entire increase.
Why This Matters: The reported increase is particularly relevant when considered alongside Google’s threat research and the UK regulators’ concerns about remediation speed.
Cyber risk is affected by both the number of opportunities attackers can exploit and the effectiveness of the controls intended to stop them.
If attackers can identify weaknesses faster while organizations continue to operate with lengthy patching and response cycles, the period during which those weaknesses can be exploited becomes more consequential.
Underwriters may need to distinguish organizations that merely possess security products from those that can respond to an incident effectively.
NICB Warns AI Is Becoming a Tool for Organized Insurance Fraud
The National Insurance Crime Bureau published new intelligence on November 16 describing the growing use of artificial intelligence in insurance fraud.
NICB reports that mentions of AI in questionable insurance claims increased by more than 1,400% between 2025 and the middle of 2026.
That figure requires careful interpretation. It measures references to AI within questionable claims, not a 1,400% increase in confirmed AI-generated fraud.
The underlying threat assessment is available only to NICB members, limiting independent examination of the methodology and baseline counts.
The organization identifies several emerging techniques, including fabricated accident photographs, altered medical documentation, synthetic identities, and identity-related fraud involving vehicle financing and cargo transportation.
NICB also reports identifying more than 1,600 transnational, interstate, and state-level organized crime rings in 2025. That is a separate measure of organized fraud activity and should not be attributed entirely to AI.
The broader concern is that widely available AI tools can make fraudulent documentation and identities less expensive to produce and more convincing.
Why This Matters: Insurance fraud has always involved an information problem. Insurers must determine whether the evidence supporting a claim accurately represents an event that occurred.
Generative AI makes that determination more difficult when convincing photographs, documents, and identities can be produced without corresponding real-world evidence.
The implications extend beyond fraud detection software.
Claims organizations may need stronger evidence authentication, more reliable records of document origins, and improved methods for identifying coordinated activity across claims and organizations.
AI can accelerate claims handling, but it may also increase the burden of establishing that the information being processed is genuine.
Marsh Says Digital Placement Can Reduce Weeks of Processing to Hours
Marsh’s third-quarter results offer a contrasting example of AI and digital technology improving established insurance operations.
During the company’s October 15 earnings call, management reported that 93% of its London wholesale trading was digital.
Marsh management also said Broker WorkBench reduced placement processing in its UK wholesale business from two to four weeks to a couple of hours. At its September launch, Marsh had projected a reduction to days or hours.
Launched in September, Broker WorkBench combines digital trading workflows, data processing, and AI-assisted capabilities to support placement requests, negotiations, and binding activities in the London specialty insurance market.
The platform allows brokers to contact markets, negotiate terms, and bind contracts without repeatedly re-entering information or passing submissions through email.
The figures are management-reported results of digital workflow changes that include AI, rather than an isolated measurement of AI’s contribution.
The processing metric should not be confused with the total time required to negotiate and bind every kind of wholesale placement.
Why This Matters: Much of the discussion about insurance AI focuses on individual tasks, such as drafting correspondence, extracting submission information, or answering underwriting questions.
Marsh’s example illustrates the potentially larger benefit of changing how an entire transaction moves through the organization.
Reducing repetitive entry, unnecessary transfers, and manual coordination can produce improvements that individual productivity tools cannot achieve by themselves.
The distinguishing feature may be less about the intelligence of a particular model and more about the way technology, information, and authority are organized across the placement process.
California Begins New Review of Consumer Protections for Insurance AI
California Insurance Commissioner Ricardo Lara announced a new initiative on November 13 to examine safeguards governing insurers’ use of artificial intelligence.
The California Department of Insurance is conducting a Responsible Use of Artificial Intelligence survey and organizing a public workshop to gather information about current practices and potential consumer protections.
The initiative addresses transparency, fairness, privacy, human oversight, accountability, and insurers’ responsibility for technologies supplied by third parties.
The department is seeking information about how insurers are deploying AI, the governance programs surrounding those systems, and their potential effects on consumers.
The survey includes a representative sample of admitted insurance companies or groups with at least $250 million in California premiums during 2025, along with selected InsurTech companies and insurers operating through managing general agents.
Survey responses are due December 30. A virtual public workshop is scheduled for December 15, with written workshop comments due December 12.
No new AI regulation was adopted through this announcement. California intends to use the information gathered to inform possible future regulatory and policy proposals.
Why This Matters: California’s initiative reflects the continuing evolution of state insurance AI oversight beyond the NAIC’s common regulatory framework.
While the NAIC supplement provides a resource for supervisory examination, California is gathering evidence that could inform its own consumer-protection requirements.
For insurers operating across multiple states, the question will be how these approaches interact as individual jurisdictions develop their expectations.
The survey may help reveal how third-party systems influence decisions affecting policyholders. Independent agents also have an interest in whether consumers can understand and challenge decisions made or supported by those systems.
Sources
- NAIC, Big Data and Artificial Intelligence Working Group Meeting Summary, November 16, 2026.
- NAIC, Innovation, Cybersecurity, and Technology Committee Meeting Summary, November 17, 2026.
- NAIC, AI Risk Evaluation Supplement v7.0, Copy Prepared for Adoption, November 2026.
- Bank of England, PRA and FCA, Effective Practices for AI-Related Cyber Risk, November 19, 2026.
- FCA, UK Financial Watchdogs Clarify Cyber Expectations, November 19, 2026.
- Google Threat Intelligence Group, Agentic Threat Activity Crosses a Significant Threshold, November 16, 2026.
- Reuters, AI Advances Coincide With Sharp Rise in Cyber Incidents, Marsh Data Shows, November 17, 2026.
- National Insurance Crime Bureau, Emerging Fraud Trends in 2026 and Beyond, November 16, 2026.
- Marsh, Third-Quarter 2026 Earnings Call Transcript, October 15, 2026.
- Marsh, Broker WorkBench Launch, September 22, 2026.
- California Department of Insurance, Consumer Safeguards for AI in Insurance, November 13, 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.
