Your weekly analysis of AI developments in insurance.
AI’s Productivity Paradox May Be Ending. What That Means for Insurance Leaders Who Waited.
In 1987, economist Robert Solow famously observed that computers were showing up everywhere except in the productivity statistics. The paradox named after him lingered for years — then, in the mid-1990s, productivity exploded. Tech remade the economy. Investors who waited too long paid for it.
Fast Company reported this week that the evidence is mounting that AI may be at exactly that inflection point. The data is coming from the most credible sources available: enterprise earnings reports. Alphabet says AI lifted core Search revenue by 19% and Google Cloud revenue by 63%, with AI-driven revenue from large clients up 800% year over year. Microsoft says its AI business is now running at an annualized revenue rate of $37 billion. Deloitte’s research finds that most companies that have adopted AI are seeing ROI, and nearly a quarter are reporting gains of 30% or more.
The Fast Company analysis draws the parallel carefully: the original Solow Paradox resolved not when better computers arrived, but when companies finally learned how to use the ones they had — building the infrastructure and processes to extract real value from the technology. The enterprise adoption numbers now appearing in earnings reports suggest that moment may be arriving for AI.
Why This Matters for Insurance:
Insurance executives who have been watching AI’s hype cycle with appropriate skepticism may be watching something else now: the hype becoming revenue. The carriers and agencies that moved from pilot to production already have a head start on the process learning that drove the productivity revolution of the 1990s. The ones still waiting are not avoiding risk — they are accumulating it. If Solow’s Paradox is resolving, the time to build internal competency is before the inflection point, not after it.
India Just Gave Insurers Three Days to Prove AI Cyber Readiness. The Rest of the World Is Watching.
India’s insurance regulator issued a directive this week that gave every insurer in the country until today — Friday, May 22 — to submit a formal Action Taken Report on their readiness for frontier-AI cyber threats. Insurance Business Asia reported that the directive from the Insurance Regulatory and Development Authority of India requires carriers to evaluate their current posture and detail the preventive, detective, and responsive security measures they are putting in place specifically in relation to AI-driven threats — not just generic cyber risk.
The compressed timeline is the signal. A multi-month assessment process compressed into days is not normal regulatory cadence. It reflects genuine urgency. The IRDAI’s revised Information and Cyber Security Guidelines, already issued in April, require insurers to notify regulators within six hours of a cyber incident, maintain continuous system monitoring, and retain log data for 180 days. The new directive adds a frontier-AI layer on top of that framework.
The geographic reach is limited for now, but the directional signal is not. Hong Kong’s Insurance Authority, Singapore’s Monetary Authority, and Japan’s FSA are all actively developing similar frameworks. How Indian insurers respond to Friday’s ATR requirements — and how IRDAI calibrates its review — will be read closely in those markets. For U.S. and European carriers, the IRDAI action is a preview of where NAIC and state-level regulators are eventually heading.
Why This Matters for Insurance:
Cyber underwriters have a direct stake in this story. If major insurance markets begin requiring carriers to formally attest to AI-specific cyber readiness, the quality of that attestation becomes a legitimate underwriting question for cyber insurers writing those carriers as risks. The India deadline also exposes what analysts already know: many insurers globally still operate on legacy IT estates that struggle to detect, let alone respond to, AI-driven attacks. The gap between regulatory expectation and operational reality is not theoretical. It is the kind of gap that produces both coverage claims and regulatory enforcement actions.
Markel Is Investing in AI-Ready Underwriting Infrastructure. The Hyperexponential Partnership Explains Why.
Markel International announced a partnership this week with hyperexponential to modernize and digitize underwriting workflows in Canada, as part of a broader push to build AI-ready underwriting infrastructure across its operations. The partnership is notable because hyperexponential’s platform is built specifically for specialty and complex commercial lines — the kind of underwriting that depends on pricing flexibility, rapid model iteration, and clean data capture.
The timing connects to a broader pattern at Markel. The company has been investing in technology infrastructure in parallel with managing pressure on combined ratios and addressing shareholder governance questions. The hyperexponential rollout is most relevant to investors and competitors watching whether those technology investments translate into pricing precision and underwriting consistency — which is where AI actually shows up in a combined ratio.
Markel also expanded its ProSolutions portfolio this week with bundled professional, cyber, media, and general liability offerings for content creators and entertainment businesses. The two moves together reflect a carrier that is simultaneously building AI-ready pricing infrastructure and moving into coverage segments where the risk profile is evolving faster than traditional actuarial frameworks.
Why This Matters for Insurance:
Specialty carriers that invest in AI-ready pricing infrastructure now are building an advantage that compounds over time. A carrier running hyperexponential or a comparable platform captures structured data on every underwriting decision — data that trains better models, supports more defensible rate filings, and surfaces portfolio-level trends faster than carriers still running pricing in spreadsheets. Markel’s investment is worth watching not as a vendor story, but as a signal about where competitive advantage in specialty underwriting is being built.
The AI vs. Automation Distinction Is Becoming Operationally Consequential for Agencies.
A piece published this week by Coverager draws a distinction that matters practically for agency operations: automation and AI are not the same thing, and deploying the wrong one for a given task wastes both money and opportunity.
The argument is straightforward. Rules-based automation — renewal reminders, routing confirmations, structured data transfers — works best when inputs are predictable and consistent. It is reliable, auditable, and cheap to operate. Where it breaks down is at the front of most real agency workflows: the unstructured email, the ambiguous client request, the document that doesn’t match the template. That is where AI adds value — interpreting intent, classifying requests, and handing structured outputs to the automation layer that follows.
The practical implication is that agencies asking “should we use AI?” may be asking the wrong question. The better question is which parts of a given workflow benefit from AI’s interpretive capacity, and which parts benefit from automation’s reliability and cost efficiency. Vertafore’s 2026 trends research found that fewer than a quarter of agency professionals believe AI will transform their everyday work this year. But that number likely reflects confusion about what AI is being asked to do, not skepticism about productivity gains from the right deployment.
Why This Matters for Insurance:
Carriers that depend on independent distribution have a stake in how their agencies answer this question. An agency that deploys AI at the intake layer and automation at the workflow layer processes submissions faster, makes fewer data errors, and surfaces coverage gaps more consistently. That has direct implications for submission quality, renewal retention, and the accuracy of data flowing back to carriers for pricing and portfolio management. Agency AI adoption is not just an agency operations story — it is a data quality story for the carriers receiving that work.
From the AI World: Altman Wins, a New Anthropic Model Raises Cyber Questions, and Salesforce’s $300 Million Bet
The Musk v. Altman verdict.
After three weeks of trial testimony in Oakland, California, a nine-member federal jury took less than two hours on Monday to dismiss all of Elon Musk’s claims against OpenAI and Sam Altman. The jury found that Musk had waited beyond the statute of limitations to file his lawsuit, which alleged that Altman and co-founder Greg Brockman had unlawfully enriched themselves by converting OpenAI from a nonprofit to a for-profit entity. The jury also rejected Musk’s claim that Microsoft aided and abetted the alleged breach of duty. Judge Yvonne Gonzalez Rogers accepted the finding and dismissed the case. Musk called the ruling a “calendar technicality” and said he would appeal.
The verdict resolved the procedural question cleanly but left the substantive one open: the jury never reached the merits of whether OpenAI’s for-profit conversion violated the organization’s founding mission. That question will likely resurface in Musk’s appeal — and it remains relevant to broader conversations about AI governance and nonprofit accountability that are playing out well beyond this courtroom.
For insurance purposes, the verdict removes a significant source of legal uncertainty around OpenAI’s corporate structure. A ruling on the merits against Altman and Brockman could have created disruption at the organization responsible for the world’s most widely deployed consumer AI product. That risk is now, at least temporarily, off the table.
Anthropic’s Mythos model and what it means for cyber insurance.
Cloudflare published a detailed account this week of Project Glasswing, an Anthropic initiative that gave selected organizations access to Claude Mythos Preview — a security-focused frontier model not available for general release. Cloudflare pointed Mythos at more than fifty of its own code repositories to see what it would find.
The results are significant for anyone writing cyber insurance. Mythos Preview represents a qualitative jump over previous models specifically in its ability to take low-severity vulnerabilities — the kind that typically sit in a security backlog indefinitely — and chain them into a single, more severe exploit. Previous models could identify isolated issues. Mythos can construct multi-stage exploit chains from vulnerability primitives. Cloudflare concluded that this capability is real, and that the model’s built-in guardrails are not consistent enough to serve as a complete safety boundary on their own.
The Project Glasswing launch partners include Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and NVIDIA — a roster that signals how seriously the largest technology organizations are treating AI-enabled offensive security capabilities. The India IRDAI directive discussed above is directly connected: the “unreleased internal model” discussion that triggered IRDAI’s emergency deadline is the Mythos Preview story.
Cyber underwriters should be tracking Project Glasswing carefully. A model that can automate multi-stage exploit chain construction changes the threat model that underlies cyber insurance pricing. The defenders using these tools are getting ahead of attackers. But the capability is not one-sided, and the underwriting frameworks built before AI-enabled exploit generation may not fully price the risk that exists today.
Salesforce will spend $300 million on Anthropic tokens this year.
Salesforce CEO Marc Benioff disclosed on the All-In podcast that his company is projected to spend $300 million on Anthropic tokens in 2026, almost entirely for coding. Benioff was direct: “These coding agents are awesome. Anthropic is awesome. I am going to probably use $300 million of Anthropic tokens this year at Salesforce. Coding. Everything’s going to be cheaper to make.” He described the shift as unlike anything in his career, saying he can implement and sell software simultaneously in a way he never could before.
The number is striking in context. Salesforce froze software engineering hiring in January 2025 after reporting AI-driven productivity gains of more than 30%. By mid-2025, Benioff said AI was handling between 30% and 50% of the company’s overall workload. Agentforce, Salesforce’s AI agent product line, has reached $800 million in annual recurring revenue. The $300 million Anthropic commitment is not a pilot — it is the operating cost of a company that has structurally reorganized around AI-assisted software development.
For insurance technology leaders, this number is worth internalizing. Salesforce is a major technology vendor for insurance carriers and agencies. A company spending that level on AI-assisted development is moving significantly faster on product development than one that is not. The gap in development velocity between AI-native and traditional software companies is becoming a factor in how quickly insurance-specific technology solutions evolve — and which vendors can keep pace with the problems insurers need to solve.
By James W. Moore
If You Found This Useful
This newsletter exists because insurance executives need trusted, sourced analysis on AI — not vendor marketing dressed up as news. If a colleague would benefit from receiving it, forward this issue or share the link.
And if you haven’t downloaded The Independent Insurance Agent’s AI Playbook yet, it’s available free at insuranceindustry.ai/ai-playbook. More than 33,000 words of practical guidance, no cost, no catch.
Sources
- Fast Company: If an obscure 1980s paradox is any guide, AI may be about to hit a huge tipping point
- Insurance Business Asia: Three days to AI-proof — India hands insurers an emergency cyber-readiness deadline
- Simply Wall St: Markel’s AI-Ready Underwriting Push and New Media Cover
- Coverager: AI vs. Automation — The Shift That’s Changing Agency Workflows
- NBC News: Jury throws out Elon Musk’s lawsuit against OpenAI and Sam Altman in less than two hours
- Cloudflare Blog: Project Glasswing — what Mythos showed us
- Anthropic: Project Glasswing
- Techloy: Marc Benioff Says Salesforce Will Spend $300 Million on Anthropic Tokens This Year
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.
