The word "agent" increasingly tags along with AI, but the term carries a lot of weight — and a lot of risk — that insurers should be mindful of.
Agentic AI can act autonomously, making it well-suited for low-risk, repetitive tasks but a compliance nightmare for anything that requires accountability. Experts in the insurance industry explain where to draw the line, and why sometimes a less capable AI is a more valuable tool.
Read our coverage below.
Intelligent workflows, not AI agents, fit insurance needs
Despite industry buzz around AI agents, insurance professionals are steering toward AI-enabled workflows that keep humans in the decision loop — as regulators may require. Fully autonomous systems create accountability gaps in an industry where claim denials, underwriting actions and risk assessments must be explainable and auditable. Rather than deploying autonomous tools that replace employee judgment, carriers and risk managers are building configurable workflow steps — summarizing documents, extracting data, flagging gaps, drafting communications — that accelerate processes without surrendering oversight. Flexible, modular AI architectures allow different teams to assemble capabilities around their specific pain points instead of forcing uniform, prebuilt solutions.
Read more:
Agentic AI will split insurance market, MACH Alliance warns
Insurers running isolated AI pilots risk a false sense of progress, according to MACH Alliance research showing carriers on composable, modern architectures deploy AI faster and achieve stronger operational and commercial results than those on fragmented legacy systems. The divide sharpens as agentic AI — systems that reason, coordinate and act continuously across the enterprise — renders siloed architectures a structural liability. Accelerating flawed processes with AI only locks in existing limitations. The strategic priority is redesigning operating models before layering autonomous systems onto them, ensuring data, workflows and platforms can support real-time coordination across underwriting, claims, fraud detection and customer servicing.
Read more:
AI speeds claims routing as 2026 hurricane season outlook stays calm
Despite forecasts of a below-average 2026 North Atlantic hurricane season — six to 16 named storms, three to nine hurricanes — a single landfalling storm can still produce billion-dollar losses, according to Allianz's hurricane season outlook. Liberty Mutual is already using AI to route claims to the right adjusters faster, reducing reassignments and enabling real-time customer support. Rapid intensification events are growing more frequent, making early exposure assessment critical. Allianz analysts stress that preparedness planning must account for asset vulnerability and exposure concentration, not storm count — history shows below-average seasons rarely guarantee below-average losses.
Read more:
Federated AI architecture lets carriers scale claims automation
Agentic AI can compress claims cycle times from more than a week to one or two days — but compliance barriers, not strategy gaps, are blocking scale adoption. The architectural answer is a federated compute layer that queries data in place rather than consolidating or exporting it, keeping regulated policyholder data within compliant environments while allowing foundation models to reason across claims systems. A three-phase execution model — governance planning, standardization of reusable infrastructure and team-level enablement — eliminates the need to rebuild compliance frameworks for each new use case. Once the first agentic workflow clears security and access review, subsequent deployments compound in speed and efficiency.
Read more:
UnitedHealth bets $3B on AI to cut costs, rebuild trust
UnitedHealth Group's $3 billion AI investment in 2026/27 — yielding a claimed 2-to-1 return — offers a benchmark for health insurers weighing technology spending against operational efficiency. The insurer projects nearly $1 billion in operating cost reductions this year, targeting the $80 billion annual administrative burden across insurers and providers, per Morgan Stanley. With 99% of its 1,000-plus AI applications focused on administrative tasks, UnitedHealth is deliberately avoiding diagnostic use — a meaningful distinction for compliance officers navigating regulatory scrutiny. An internal ethics review board clears all new AI applications. Given that 69% of Americans distrust businesses' AI use and prior authorization litigation is ongoing, transparency about governance structures and patient-facing outcomes is essential to sustaining public and regulatory credibility.
Read more:
Manulife hits $300M toward its $1B AI value goal
Manulife has captured roughly $300 million of its $1 billion AI value target for 2025-2027, driven by personalized customer offers, fraud detection, software development cost reductions and AI-assisted underwriting tools including John Hancock's Quick Quote and Canada's MAUDE engine. The insurer extended Microsoft Copilot access to all 38,000 employees and built a 5,000-member internal AI Community of Practice to accelerate adoption. Its approach — embedding AI across all business functions before scaling governance and operations — offers a replicable framework. Executives received dedicated AI training, and leadership revisits AI strategy quarterly, underscoring that one-time workforce training is insufficient for sustained ROI.
Read more:
AI uncertainty drives insurance M&A down to $30B in early 2026
Insurance M&A volume fell to 191 transactions valued at nearly $30 billion in the six months ending May 31, 2026, down from 207 deals worth $32 billion in the prior period, per PwC's midyear outlook. The culprit: investor uncertainty over whether AI will empower disruptive new entrants or strengthen incumbents through lower-cost brokerage and improved underwriting margins. Despite the dip, deal fundamentals remain solid — premium rate increases, tighter underwriting and improved profitability continue drawing capital. Carriers and brokers actively investing in AI-enabled underwriting, claims and workflow automation may see improved valuations and position themselves favorably as M&A scrutiny intensifies.
Read more:
WTW: Match AI deployment to function, keep humans in loop
Generative AI is ill-suited for risk prediction — querying the same model 50 times yields 50 different answers, making explainability impossible under regulatory scrutiny, according to WTW global insurtech innovation leader Magdalena Ramada. Traditional machine learning and GLMs remain the appropriate tools for predictive modeling, augmented by AI flagging anomalies. Agentic systems require continuous human shaping to remain viable, and complex actuarial functions such as reserving cannot be fully automated. AI adoption will accelerate fastest in customer service and distribution; actuarial transformation will be incremental. Executives claiming headcount reductions are AI-driven will likely need to rehire subject-matter experts as systems scale.
Read more:
AI black box risks threaten insurers' compliance, costs
Insurers building AI systems without a dedicated lifecycle management function risk deploying unscalable, error-prone tools that are costly to audit and difficult to fix, according to WTW's Magdalena Ramada. Regulatory exposure is immediate: Black box models cannot meet explainability requirements for pricing, reserving and underwriting decisions. Concentration risk compounds the problem — with the industry converging on a handful of foundation model providers, systemic fragility is rising across the sector. Licensing and token-pricing changes already announced by major AI vendors signal future cost increases, a dynamic Ramada compares to third-party data dependency that has existed for 15 years. Architectures that limit model reliance to targeted use cases reduce exposure.
Read more:
Four-phase AI framework to decouple growth from rising costs
Commercial insurers can scale policy count without proportional expense growth by executing a four-phase AI strategy: digitizing unstructured submissions into insurer-specific risk data; aggregating that data across all communications into a complete risk picture; automating end-to-end workflows including routing, triage and straight-through processing; and embedding those workflows into broker-facing systems. The final phase carries the highest competitive weight — brokers place risk with carriers that reduce friction, and insurers with real-time submission visibility and proactive appetite signaling are structurally harder to displace. Legacy barriers like siloed data are increasingly solvable through modern integration, narrowing the window for delayed adoption.
Read more:
This roundup was created with AI assistance. A Digital Insurance editor reviewed each item before publication.
