Insurers double down on AI—but trust sets the pace
The gist
Insurers are pouring money into AI, but trust, transparency, and human oversight—not flashy tech—are now the ticket to scale.
What to know
- By September 2026, insurers like IAG went from AI pilots to launching more models in two years than the previous six, building 150 Gen AI activators across the business.
- AI is now embedded in live operations—platforms like Capgemini's RAG chatbot, NTT DATA’s AIVista, and Verisk’s Claude connectors are reshaping claims and underwriting with real-world results.
- Governance is the bottleneck: 74% of Australian insurers use gen AI in claims, but 70% cite regulatory risk, with most still relying on in-house and manual review to avoid costly AI errors.
AI Budgets Demand Trust
Insurers are pouring record budgets into AI, but scaling hinges on rapid governance and human oversight to keep pace with regulatory demands and maintain competitive relevance.
By September 2026, the insurance industry’s AI debate had moved from experimentation to budget-backed operational scaling, with trust as the condition for going faster. In Scouting for Growth, IAG’s Willem Paling described the shift as moving from isolated proof-of-concepts to established AI products, embedded governance, and human-AI operating models built for regulated insurance environments, and said the task was to govern it, scale it, and trust it fast enough to remain relevant. The same case study said IAG had launched more AI models in the past two years than in the previous six combined and built a network of 150 Gen AI activators across the business.
That September consensus also tied larger AI commitments through 2028 to governance, transparency, and human oversight rather than unchecked automation. Insurance Business cited studies showing 80% of leaders say human-in-the-loop AI is important for business operations and that nine in 10 individuals are more likely to trust a company that uses it, while FinTech Global said the focus was on scaling AI beyond pilots without losing control or appropriate oversight; Unite.AI said, “This is what has changed in 2026,” and Aetna’s Nathan Frank said AI agents supporting over 12,000 nurses have given clinicians 90 minutes back per day.
AI Platforms Go Mainstream
Enterprise-grade AI is now powering live insurance workflows, with platforms like Capgemini, NTT DATA, and Verisk delivering measurable speed and accuracy gains across claims and underwriting.
The clearest sign that insurance AI is leaving the pilot phase is that deployment now looks like enterprise operations, not experimentation. A July warning argued that “BCBSM’s 50x token explosion is not an anomaly — it is a preview of what every enterprise will face as agentic AI moves from pilot to production,” because inference demand is rising as capacity tightens; in practice, insurers are responding with roadmaps and architectures, as Capgemini built a RAG-based intelligence chatbot, compared Mistral AI and GPT-4 with model evaluation frameworks, and produced reference architecture to measure and control business value.
What makes this shift measurable is that platforms are being embedded into live workflows with controls and visible output, not left as side experiments. Capgemini said one insurer’s travel chatbot “achieved 91% accuracy with live agent validations,” while IT Brief New Zealand reported, “NTT DATA Launches AIVista-Powered AI Platform, Promises 3X Faster Claims for Top Global Insurers”; in May 2026, Verisk launched two Model Context Protocol connectors with Anthropic for underwriting and claims that “bring Verisk data and analytics into Claude workflows,” TradingView added that Verisk’s Anthropic connectors bring underwriting and claims data into Claude workflows, XactAI users rose nearly tenfold to almost 7,000 licensees, and Verisk “released seven Core Lines Reimagine modules in the second quarter and remains on track to deliver 25 in 2026,” expanding “the company's ability to embed updated data, analytics and workflow capabilities across” core lines.
Accountability Shapes AI Limits
Insurers are locking AI behind strict governance and manual checks, treating mistakes as existential risks and making transparency non-negotiable for claims and underwriting decisions.
Governance has become the gating constraint because insurers increasingly view AI mistakes as balance-sheet, conduct, and franchise risks rather than mere technical glitches. FinTech Global reported that “the question driving boardrooms in 2026 is no longer how fast AI can be deployed but how it can be deployed responsibly,” since in insurance errors can mean “losses running into the millions”; IntellectAI warned a poorly trained model could decline profitable business, introduce pricing bias, or generate inaccurate policy language, which is why governance now spans underwriting, legal, compliance, cybersecurity, and boards, with human review, audit trails, and drift controls treated as operating necessities.
Trust and transparency are the other preconditions because insurers must be able to explain, challenge, and ultimately own consequential decisions. “People in 2026 increasingly demand to know why an AI made a decision on their behalf,” and that expectation is colliding with claims operations: Gallagher Bassett said 74% of Australian insurers are using generative AI in claims resolution, yet 70% named regulatory and compliance risk as the biggest barrier to further integration; the CSIRO and Insurance Council of Australia’s joint 2025 report, AI for Better Insurance: Enhancing Customer Outcomes amid Industry Challenges, identified priority use cases including “automated claims processing,” and CSIRO and the Insurance Council of Australia called for stronger governance around automated claims and underwriting, while Gallagher Bassett found 90% keep in-house validation teams, 68% use manual review, and 20% external verification, underscoring that accountability cannot be left ambiguous when 18% of finance leaders say a major AI-agent error might belong to no one.
