Insurers can't trust AI unless they can also defend it

The big question around AI in insurance is not whether to adopt it, but how to adopt it safely and securely. 

It's not enough to simply trust that your AI tools work properly; you have to be able to prove it to outsiders, including customers and regulators. This is a sentiment echoed by several of the thought leaders who contributed to Digital Insurance's InsureThink page in the past week. 

"Analysts need to be able to understand how an AI agent came to a certain conclusion and adjust it, as necessary," writes Maik Taro Wehmeyer, co-founder and CEO of Taktile. "For instance, we need to see what data the AI used to decide to give the green light on an insurance claim. This visibility is nonnegotiable."

Read what Wehmeyer and others said by clicking the links below:

Audit AI underwriting criteria before state AGs do

State attorneys general are ramping up scrutiny of insurers' AI-driven underwriting and claims practices, citing new privacy laws in states such as California, Colorado and Virginia alongside existing consumer protection and civil rights statutes. Pennsylvania's 2024 settlement with Geico required longer documentation windows for AI-flagged policyholders, signaling regulators' focus on transparency. Insurers face growing exposure if disclosed underwriting criteria don't match algorithmic inputs, or if automated decisions disproportionately affect protected groups without meaningful human review. Auditing AI models for these mismatches now can help preempt enforcement, says Jason Downs, a partner in Hogan Lovells Cadwalader's disputes practice group.
Read more: Is your AI underwriting on state AGs' radar?

Monitor AI override rates as claims automation arrives in 2026

Frontier AI models have crossed a reliability threshold that will drive insurance claims automation in 2026, writes Maik Taro Wehmeyer, CEO of Taktile. To deploy AI safely, insurers need a "decisioning layer" connecting AI, agents and humans, giving compliance teams visibility into decisions affecting claims payouts, fraud detection and P&L. Wehmeyer recommends tracking two metrics: approval/decline rates, which confirm alignment with business goals, and override rates, which measure how often humans correct AI decisions — a signal the system is learning and improving accuracy over time. Without these guardrails, automation only heightens regulatory and financial risk instead of improving performance.
Read more: Why 2026 is the year AI finally automates insurance claims

Modernize core systems before scaling AI, expert says

Insurers chasing AI scale often blame messy data, but the real bottleneck is outdated policy administration systems that can't keep pace with new AI agents, writes Kevin Gaut, chief technology officer at Instanda. When underwriting rules and AI layers fall out of sync, agents risk producing inconsistent decisions across a policy's lifecycle — from binding to mid-term adjustment to renewal. Before adding more AI tools, insurers and MGAs need to audit core systems, ensure ISO 27001:2022-level data security, and keep humans reviewing outputs for hallucinations. Modernizing the PAS foundation first, rather than layering AI onto legacy infrastructure, is key to sustainable scale.
Read more: Insurers' AI stalls when core systems don't scale

Share claims data to curb rising check-fraud risk

The Federal Reserve's 2024 payments study shows noncash payments hit $236.6 billion, with ACH now carrying nearly 75% of that value as checks continue a steady decline. As routine transactions go digital, the checks that remain — many tied to claims — skew toward higher-value, higher-risk payments, driving up fraud, escheatment and reconciliation costs. Pennsylvania now mandates direct deposit for injured workers by year-end, and a federal executive order is pushing agencies away from paper checks entirely. Shareen Minor, chief revenue officer, U.S. at Vitesse, argues that carriers should unify claims and finance data across TPAs and banks to flag risky payments before disbursement, cutting exposure as check volume shrinks.
Read more: To fight rising check fraud, carriers must open up their data

Modernize back-end systems to boost broker loyalty

Carriers seeking stronger broker relationships should prioritize modernizing back-end underwriting infrastructure over front-end submission tools, writes Michael Topol, co-founder and co-CEO of MGT Insurance. Legacy systems force underwriters to manually gather hundreds of data points per account, often across seven to 10 disconnected applications, slowing quote turnaround. Capgemini research cited in the piece found 61% of underwriters struggle to improve quote-to-bind conversion and 49% struggle with agent and customer satisfaction. Deploying AI to automate data gathering and validation — not underwriting judgment itself — can compress decisions from weeks to minutes, giving carriers a durable edge in broker loyalty and retention.
Read more: Insurers using AI for underwriting will increase broker loyalty

Factor wildfire smoke's lagging health risks into pricing

Wildfire smoke's fine particulate matter (PM2.5) can raise disease prevalence by 2.3% to 8.6% and boost mental health disorder rates by up to 7.8%, with effects sometimes emerging months or years after exposure, writes Kara Clark, senior research actuary at the Society of Actuaries Research Institute. Because wildfire frequency and burned acreage don't reliably predict pollution-related mortality, actuaries should move beyond event-based metrics and build long-tail climate risk into morbidity and mortality models. Incorporating lagged cardiovascular, respiratory and mental health impacts — and disparities across underserved communities — into pricing and contingency loadings can improve claims forecasting as fire seasons keep lengthening.
Read more: Why insurers should include climate risk in modeling: SOA

Close five security gaps to boost cyber-risk resilience

During leadership transitions, protect what already works rather than rushing to change security processes, since distraction invites attackers. Closing five gaps — awareness, skills, trust, implementation and certainty — is key to strengthening cyber resilience. Run phishing simulations before incidents occur, recruit talent beyond domestic markets, and apply the same rigor to cyber risk as financial due diligence during M&A. Map data flows to identify disconnected systems, and build multiple contingency scenarios instead of betting on one outcome. Near term, test incident response playbooks and prioritize patching by exploitability; long term, invest in zero-trust architecture, says Steve Durbin, chief executive of the Information Security Forum.
Read more: Building security awareness increases risk resilience

This roundup was created with AI assistance. A Digital Insurance editor reviewed each item before publication.