A New Kind of Risk
When you take out insurance for your business, you know roughly what you are insuring against. Fire, flood, theft, public liability — these are well-understood risks with decades of claims data behind them. But what happens when the thing you are insuring is an AI system that makes decisions, interacts with customers, and modifies its own behaviour over time?
That question is the subject of serious academic research. A new paper from researchers in the field of AI-native insurance develops a mathematical framework for underwriting, pricing, and designing insurance contracts for what they call “agentic AI” — autonomous AI systems that can make decisions, use tools, and interact with third-party services without direct human supervision.
This is not a hypothetical problem. Irish businesses are already deploying AI agents for customer service, data analysis, inventory management, and even contract review. If that AI makes a mistake — overcharges a customer, shares confidential information, or makes a bad business decision — who is liable?
The Insurance Framework Explained
The research proposes a risk state model that captures five things: the AI’s autonomy level, its operational authority, its permission exposure, its governance maturity, and how concentrated its dependencies are. Think of it like a driver profile for your AI. A low-autonomy AI that only makes suggestions under human supervision would get a low-risk rating, similar to a careful driver. A fully autonomous AI that can make financial decisions and interact with customer accounts would be rated much higher.
The framework then maps that risk state to things like event probabilities, loss severities, governance costs, premiums, deductibles, and coverage limits. It even accounts for things like “governance certification thresholds” — basically, proving that you have proper oversight in place before you qualify for coverage.
For an Irish business, this is not as abstract as it sounds. If you are using AI to handle customer inquiries, process orders, or manage stock levels, you already have what the researchers would call an “AI deployment.” Your existing business insurance almost certainly does not cover AI-specific liabilities.
Why This Matters Now
The insurance industry moves slowly, but it moves deliberately. When a major insurer starts offering AI-specific policies, it will be because they have done the maths. The research being published now is the foundation for those policies.
Ireland’s position as a tech hub means our businesses are often ahead of the curve in adopting AI tools. But being ahead means also being exposed to risks that insurers have not yet priced. A small Irish business using an AI booking system for appointments, or an AI tool for generating quotes, could find itself in uncharted territory if something goes wrong.
The European Union’s AI Act, which is now being implemented, will add another layer. Under the act, high-risk AI systems will face strict requirements for transparency, human oversight, and accountability. Insurance may become a requirement rather than an option.
Practical Steps for Irish Businesses
You do not need to buy AI-specific insurance today, but you should start preparing. First, audit your use of AI. Make a list of every tool or system that makes autonomous decisions. Second, talk to your insurance broker about whether your existing policy covers AI-related liabilities. Expect a blank look — but the conversation itself is useful. Third, consider what governance you have in place. Do you have a human reviewing AI decisions? Do you log AI actions for audit purposes? These are exactly the things the new insurance frameworks will look at.
AI insurance is coming. The businesses that think about it now will be the ones that get better terms when it arrives.