InsuranceIndustry.AI

An independent commercial lines agency writes $8 million in premium. At 11.5 percent commission, that is $920,000 a year for connecting clients to carriers. For decades, nobody has had to formally answer what each of those dollars is paying for. The people who will ask that question next are already building the tools to answer it themselves — and the cost curves they are seeing are collapsing faster than the commission structure assumes.

OpenAI funds an insurance underwriting startup, a reinsurer settles the first command-line AI agent payment, and a new survey finds most underwriters use AI but don’t trust it. Plus Anthropic’s most capable model yet, AI-enabled cyber threats reshaping risk, and a CEO’s case for regulating frontier models. This week’s AI Insights.

In February 2026, the insurance market moved in two directions at once. ElevenLabs went live with the first dedicated AI liability policy, backed by Lloyd’s market capacity. In the same underwriting cycle, ISO exclusions took effect, explicitly removing generative AI losses from commercial general liability coverage.
That is the market admitting the exposure is real, and the form is wrong.
AI risk is insurable. But the industry won’t price it the way it has priced everything else. History shows exactly how this works, and it has very little to do with actuarial data.

Two of the world’s largest professional services firms — PwC and KPMG — announced global alliances with Anthropic this week, putting Claude in front of a combined 300,000+ professionals and directly into client engagements at insurance carriers. Anthropic filed confidentially for an IPO at a valuation approaching $1 trillion. Bain surveyed 951 companies and found that AI budgets keep growing while returns keep falling short — and identified exactly why. New KPMG data shows 73% of insurance CEOs have named AI their top investment priority, but only one in three has seen meaningful returns. Honeycomb Insurance raised $40 million on a profitable, no-inspection AI underwriting model covering $100 billion in insured value. The White House signed an AI executive order focused on cybersecurity — not regulation — with direct implications for carriers running AI in production. And Google released Gemini 3.5, raising the speed and cost floor for agentic AI deployments. All of that, with insurance context, in this week’s AI Insights.

Deepfake fraud was alarming. Synthetic claims are a different problem entirely. The threat has evolved from a single convincing document to an entire fabricated evidence ecosystem designed not to fool a human reviewer, but to survive automated claims workflow validation. In this follow-up to our January analysis, we examine synthetic claims, camera injection attacks, the collapse of passive trust, and what forward-looking insurers are doing about it.

Three states have enacted laws governing how health insurers can use AI in claims decisions — and the frameworks are meaningfully different from each other. A specialty insurance market is scrambling to define what it’s actually covering as D&O and E&O carriers grapple with AI washing, silent AI risk, and boards that can’t define “reasonable oversight.” An AI-first specialty insurer runs 70% of its business autonomously and has lessons for carriers still fighting legacy infrastructure. Thrive Capital and Sequoia just co-led a $46 million Series B in an AI insurance operations platform that has already completed 250,000 workflows. AllianceBernstein argues that AI is repricing the liabilities side of the insurer balance sheet, not just operations. And Microsoft’s AI chief is predicting full white-collar automation within 18 months — here’s what the counterevidence says and why the timeline ambiguity doesn’t change the strategic math for insurance leaders.

Every insurance carrier has an org chart. But here’s what most executives don’t see: that org chart isn’t just a management tool. It’s a map of what it once cost to move information and expertise through the enterprise.
AI is changing those costs. And organizational structures built around the old economics don’t automatically update. They persist. And they become expensive.
This isn’t an argument about tools or technology. It’s an argument about architecture — and why the carriers that will lead the next decade aren’t asking “how do we adopt AI?” They’re asking a harder question: was our organization designed for a world that just ended?

India gave insurers three days to prove AI cyber readiness. A federal jury took less than two hours to end the Musk v. Altman trial. Salesforce is spending $300 million on Anthropic tokens this year — almost entirely for coding. Cloudflare tested an unreleased Anthropic security model on fifty of its own code repositories, and the results have direct implications for cyber underwriting. Markel is investing in AI-ready pricing infrastructure through a new hyperexponential partnership. And new data suggests AI’s long-awaited productivity tipping point may finally be arriving. This week’s issue covers all of it, plus what the AI vs. automation distinction means for agency operations — and for the carriers receiving that work.

BCG reports that 74% of large-scale insurance IT transformations fail. The other approaches — lift-and-shift, big bang migration, phased migration — all share the same fatal constraint: they require the legacy carrier to transform itself from the inside.
The AI Sidecar Strategy is the sixth path. It builds the replacement outside the legacy architecture entirely, using the data advantage legacy carriers already hold.
This white paper defines the strategy, maps the migration mechanics, names the failure modes, and makes the board-level case for why inaction is the riskier position.
The only thing more expensive than building a sidecar is competing against one.

Iowa just made AI-only claim denials illegal. The Society of Actuaries named AI the top emerging risk across all insurance sectors — including P&C and life — for the first time. Outmarket AI closed a $17 million Series A with a reported 65% reduction in E&O exposure across its brokerage deployments. A Nobel-winning MIT economist published research showing U.S. firms have historically used automation to target wage earners rather than maximize productivity — with implications for how carriers should think about AI workforce strategy. And a practitioner from Arthur J. Gallagher made the case that generic AI safety tools cannot address insurance’s unique compliance and accuracy requirements. This week’s issue covers all of it, plus a recap of what happened when Dario Amodei and Jamie Dimon shared a stage to discuss AI in financial services — and why the Verisk connector announcement buried in the product news matters for insurance carriers.