Three insurance-technology stories from the week of October 19–26 are worth more than a headline: Chubb’s CEO saying the industry has not yet priced AI liability, regulators asking insurers directly about agentic AI, and a $27 million bet that embedded insurance needs its own kind of carrier. Here is what happened and why we think each one matters.
1. Chubb says AI liability is the exposure nobody has underwritten yet
What happened: On Chubb’s third-quarter earnings call, reported October 22, CEO Evan Greenberg described liability from AI as a growing, emerging exposure that casualty underwriting has not yet accounted for. He said he expects the industry to respond with coverage clarifications, exclusions and separate AI liability products, followed by underwriting and pricing. He also suggested certification standards and best practices for how companies use AI and AI agents. (Insurance Journal)
Our take: When the CEO of one of the world’s largest commercial insurers says the industry has to get its arms around a risk, the policy wording usually follows. That puts pressure on two groups at once. Underwriters need a way to find and measure AI exposure in an account. Insurers also need to show that their own AI use is controlled, because they are about to start asking their clients the same questions.
What we expect to see over the next few renewal cycles:
- AI questions added to commercial applications, similar to how cyber questionnaires spread a decade ago
- Exclusions and endorsements that clarify whether existing liability forms cover AI-caused harm
- Standalone AI liability products, priced on the controls an insured can document
- Demand for third-party certification of AI governance, which some insurers may make a condition of coverage
If you build underwriting tools, this is a data problem you can start solving now: how to capture an insured’s AI use in a form that can be underwritten and priced.
2. Regulators want to know what your AI agents can do on their own
What happened: The NAIC’s Big Data and Artificial Intelligence (H) Working Group released version 6.0 of its AI Risk Evaluation Supplement for comment. The draft adds a question asking whether a use case involves agentic AI, what it is meant to do, and whether it can act without human approval. It also asks about testing for proxy effects, and it lets regulators leave traditional generalized linear models (GLMs) out of the more detailed questions. Comments are due October 30, with a public call scheduled for November 2 and a session at the NAIC Fall National Meeting on November 16. (InsureAI Wire, NAIC draft)
Our take: This is the first time the supplement asks about agentic AI by name, and the question is a practical one: what can the system do without a person signing off? Many carriers are piloting agents for intake, first notice of loss and underwriting triage without a clear written answer to that question. Narrowing the scope for GLMs is also a sensible signal. Regulators are focusing their attention on newer, less transparent models, rather than on rating techniques they have reviewed for decades.
Steps worth taking before the next market conduct exam:
- Inventory every AI use case and tag which ones involve generative or agentic AI
- For each agent, write down what it can do without human approval, and where a person must step in
- Document how you test for proxy discrimination, even if the answer is “not yet”
- Ask your vendors for the same information about the agents built into their products
3. Slingshot’s $27M seed: embedded insurance wants a carrier built for it
What happened: On October 22, Slingshot Insurance came out of stealth with a $27 million seed round co-led by Rally Ventures and Skyward Group. Led by Ethan Genteman, formerly chief actuarial officer at Palomar, the company describes itself as an AI-native insurer for embedded insurance. It serves MGAs, vertical software platforms and distributors in areas such as amateur sports, travel and gig work. (Slingshot announcement)
Our take: The numbers behind embedded insurance are very different from traditional commercial lines. Slingshot’s early customer, Vertical Insure, reports more than 750,000 customers with an average premium of about $25. At that size, a policy cannot carry much underwriting or servicing cost, so automation has to come first. A large seed round for a carrier designed around that reality suggests investors see the capacity side of embedded insurance as a bottleneck, not just the distribution side.
Questions for MGAs and platforms selling insurance at the point of sale:
- Can your current carrier partner price, bind and change products as quickly as your software ships?
- What does a policy cost you to issue and service, and how much of that is manual work?
- Who owns the customer data, and can you take it with you if you change carriers?
- How quickly can a new product get filed and approved in the states you need?
Also from the week: Donegal Insurance Group expanded its use of ZestyAI’s property-risk models across its personal lines portfolio (ZestyAI), and OPTIS Partners reported that insurance agency M&A deals fell 27% in the third quarter (Insurance Journal).