Simon Glairy is a distinguished authority in the insurance and Insurtech sectors, renowned for his strategic focus on risk management and the integration of artificial intelligence into traditional underwriting frameworks. As the industry grapples with the transition from legacy systems to autonomous tools, Glairy provides a necessary bridge between technical innovation and regulatory compliance. His expertise is particularly vital today, as insurance carriers face the daunting “black box” problem where the lack of transparency in AI models can lead to significant legal and operational setbacks. By examining the shifting landscape of internal software development and the risks of third-party data dependency, Glairy offers a roadmap for insurers looking to navigate the complexities of modern risk assessment.
In this discussion, we explore the fundamental challenges insurers face when pivoting from being software consumers to internal developers, particularly the need for new operating models that manage the entire lifecycle of AI tools. We delve into the critical issue of explainability in “black box” systems, which poses a direct threat to the legal requirement for transparent pricing and underwriting. The conversation also highlights the systemic fragility caused by the industry’s heavy reliance on a small handful of foundation model providers and the financial risks associated with third-party data sources, such as wearable technology. Finally, we address the cultural shift required to prevent employee disengagement and productivity loss as AI becomes a daily fixture in the insurance workplace.
How has the traditional insurance operating model struggled to adapt to the new reality of building and maintaining internal AI tools?
The primary hurdle is that most insurance companies have historically functioned as consumers of technology rather than creators of it. For decades, the industry was built around buying off-the-shelf software, but the move toward AI requires a new paradigm where insurers must manage the entire lifecycle of internal tools. This shift demands a specialized function to ensure these tools comply with strict regulations while also maintaining rigorous information security and cyber security standards. Without this dedicated oversight, adoption increases in a fragmented way, leading to various agents that evolve inconsistently and create a massive monitoring burden. It is a fundamental change in business identity, requiring companies to generate internal solutions that they previously would have simply outsourced.
What are the most significant risks you see when insurance companies build AI systems without a deep understanding of the underlying technology?
One of the most pressing dangers is that we are currently building systems that simply do not scale because the foundation is flawed or misunderstood. Because the technology is so complex, insurers are inadvertently baking in errors that are either incredibly difficult or prohibitively expensive to spot until they cause a major failure. There is also a missed opportunity where companies fail to combine generative AI and agentic systems with more cost-effective, industry-adequate methodologies. This lack of a hybrid approach leads to a “black box” scenario where the lack of transparency makes it nearly impossible to fix issues efficiently. When you don’t understand the machine you’ve built, you end up with a tool that is more of a liability than an asset for the long-term health of the firm.
In what ways does the “black box” nature of modern AI conflict with the legal and regulatory obligations of a standard insurance carrier?
The conflict is fundamental because the insurance industry is built on the bedrock of explainability. Legally, insurers must be able to provide clear, data-driven justifications for their pricing, reserving, and underwriting decisions to regulators at a moment’s notice. A black box model, by its very definition, cannot support that requirement because the architecture obscures the “why” behind any given output. If a carrier cannot explain why a specific premium was set or why a claim was denied because the AI’s reasoning is hidden, they are opening themselves up to massive regulatory fines and litigation. This is why the specific architecture of the AI matters so much; insurers cannot afford to trade transparency for raw predictive power.
Beyond the technology itself, how is the rapid adoption of AI impacting the workforce and the internal culture of these firms?
There is a significant amount of fear and anxiety regarding how this technology will change daily roles, and if not managed, it creates a toxic cultural gap. We are seeing a divide where the most efficient and high-performing talent might embrace the tools, while others feel left behind, leading to widespread disengagement. When employees don’t understand the technology they are forced to interact with every day, their productivity naturally drops as they lose confidence in their own workflows. Management must navigate this cultural change carefully to ensure that the human element remains integrated and motivated. Without a strategy to bridge this talent gap, the very tools meant to increase efficiency could actually end up stalling the company’s output.
What kind of systemic fragility is created when the entire insurance industry relies on a very small number of AI foundation model providers?
We are currently seeing a dangerous concentration of risk where the “brains” of the industry’s AI systems are held by only about five major companies. When everyone is building on the same small group of providers, such as Microsoft or Anthropic, it creates a systemic fragility that goes far beyond the exposure of any single insurance firm. If one of these providers changes an update or their model breaks, it can cause a ripple effect that disrupts the entire sector simultaneously. Insurers are desperately trying to figure out how to remain technology-agnostic to avoid this, but the blistering pace of evolution makes that choice extremely difficult. You are essentially building your future on an ecosystem you do not control, which is a high-complexity environment with very little room for error.
How do third-party data dependencies, like those from wearable devices, complicate the long-term financial viability of AI-driven insurance products?
The risk of rising costs is very real, as we have seen over the last 15 years while innovating on data that the insurance companies do not actually own. For example, if a life insurance product is heavily dependent on data from an Apple Watch to set its rates, the insurer is at the mercy of Apple’s pricing strategy for that data. If the cost of those licenses or tokens increases sharply, the insurance product may suddenly cease to be profitable or even functional in its current form. It is a recurring problem where we use methodologies that rely on external data to “enrich” our models, but we don’t own the source. To navigate this, insurers must architect their systems to minimize reliance on these high-cost models and only leverage them where they provide the most critical value.
What is your forecast for the future of AI in the insurance sector over the next decade?
I anticipate a significant “white-boxing” movement where the industry moves away from opaque, third-party systems toward more transparent and proprietary architectures. Over the next decade, the insurers who survive will be those who successfully treated AI as a core competency rather than a plug-and-play utility, effectively managing the high-speed evolution of the technology while maintaining regulatory clarity. We will likely see a winnowing of the market, where companies that failed to address the systemic fragility of the “Big Five” providers face massive operational shocks. Ultimately, the winners will be those who balanced the raw speed of generative AI with the cold, hard requirements of explainable risk management.
