Getting insurance for frontier tech is a total nightmare right

I was talking to one of our hardware partners last week who was losing sleep because they couldn't get coverage for a pilot program. They're building some incredible autonomous systems, but every traditional broker they spoke to just hit them with a flat "no." It wasn't even because the tech was too risky; one guy was literally denied because his team uses CAD software in their workflow. It sounds ridiculous, but that is the current state of the industry.

The core issue is that insurance companies aren't actually built for the AI era. Most of them don't even write their own policies from scratch. They license standard forms from an organization called ISO and then slap on these "add-on" clauses to tweak the risk. Back in January, ISO released clauses that basically void coverage for any losses arising from AI used anywhere in a process.

The problem is that these clauses are incredibly vague. The current definition of generative AI focuses on systems that produce text, images, or code. But what happens when you have a robot control model that produces motor commands? Is that "generative"? Underwriters are giving totally different answers every time you ask. This lack of clarity is a massive bottleneck for any company trying to deploy an LLM agent or a physical robotics system in a real-world environment.

I've been following a new startup called Risklytics (they're in the YC S26 batch) that is actually trying to solve this specific deployment hurdle. They aren't just another brokerage; they are building a specialized workflow for companies working with drones, satellites, and autonomous systems.

Here is how they are tackling the mess:

• Structured Application: Instead of forcing you to fill out a generic form designed for a bakery or a law firm, they use a six-step process that asks what you are actually building. This generates a structured version that insurers actually understand, without including irrelevant fluff that triggers red flags.
• The AI Clause Map: This is the most interesting part. They have a way to see which insurers are aggressively adding those "AI-voiding" clauses and which ones are actually willing to cover frontier tech.
• Transparency on Denials: If an insurer says no, they don't just leave you hanging. They show you the specific reason and—more importantly—what would need to change for them to say yes.

It’s a pretty clever way to handle the "black box" problem of insurance underwriting. They also don't charge brokerage fees; they just take a commission from the insurer, which is standard.

If your team is moving from R&D into a real-world pilot, you're going to hit this wall. Whether it's a bridge collision system or a robotics rental service, the "safe" answer from an underwriter is almost always "no" unless you know exactly how to frame your tech. This kind of specialized AI workflow for compliance and risk is going to be mandatory if we want to see more autonomous systems actually hitting the streets.