Simon Glairy is a preeminent figure in the Insurtech space, recognized for his deep expertise in bridging the gap between traditional risk management and the frontier of artificial intelligence. As the industry stands on the precipice of a massive shift in how policies are priced and claims are handled, Glairy has become a leading advocate for a disciplined approach to algorithmic governance. In this conversation, we explore the evolving landscape of insurance, focusing on the move toward autonomous AI agents, the technical debt inherent in legacy systems, and the necessity of building “trust engineering” into the very architecture of modern insurance platforms. We delve into how the industry can navigate stringent regulations like the EU AI Act while turning transparency and auditability into a distinct commercial advantage.
As we see artificial intelligence moving from a supportive tool to an agentic model that can initiate workflows and make decisions independently, how is this transition fundamentally changing the risk profile for modern insurers?
The shift from AI as a passive assistant to an active agent represents a massive leap in operational complexity that many insurers are only beginning to grasp. When AI merely supports an underwriter or a claims handler, there is always a human filter acting as the final safeguard for consistency and fairness. However, as we move toward agentic models that orchestrate workflows and progress decisions in real time across live operational environments, that human buffer begins to thin. This creates a scenario where the system is making hand-offs and shaping customer outcomes with significantly less direct oversight, which can carry heavy financial and personal consequences if the logic is flawed. We are no longer just looking at a tool that suggests a price; we are looking at an integrated fabric of decision-making where the “control plane” must be architected to handle conflicting recommendations and autonomous boundary-setting from the very start.
You have argued that by 2026, the insurance industry will be defined by a new discipline called “trust engineering.” Why is it no longer sufficient for trust to exist as a brand promise or a regulatory response?
For decades, trust in insurance was a “soft” concept, built on the reputation of the brand, the clarity of policy wording, and the responsiveness of customer service teams. In an AI-enabled market, that is no longer enough because the speed and scale of automated decisioning move far faster than traditional human-led oversight can track. Trust engineering means that fairness, compliance, and transparency are no longer just goals for the marketing department; they are technical requirements built into the data flows and operating models of the business. By 2026, the winners will be those who can provide hard evidence of how a specific decision was reached, using systems that are designed for auditability from the first line of code. If an insurer cannot prove their AI models are treating customers fairly in real time, they will lose the trust of regulators, partners, and the very customers they aim to protect.
The EU AI Act specifically labels AI systems used for risk assessment in life and health insurance as “high-risk.” How does this classification change the way companies must approach their internal governance and record-keeping?
This classification by the EU AI Act, alongside guidance from the European Insurance and Occupational Pensions Authority, elevates AI governance from a back-office compliance task to a core operational priority. For life and health insurers, this means there is now a strict legal obligation to ensure high data quality, robust cybersecurity, and, perhaps most importantly, explainability for every automated decision. It is no longer acceptable to have “black-box” models where even the data scientists struggle to explain why a certain premium was set or why a claim was triaged for additional review. Insurers must now maintain meticulous records of data inputs, model outputs, and even the human interventions or overrides that occurred during the process. This shift forces the industry to confront whether they actually have the technical infrastructure to support such a high level of transparency and control.
Many insurance organizations are still operating on fragmented legacy systems where data sits in silos. What are the practical dangers of trying to deploy sophisticated AI on top of these older technology estates?
The most significant danger is that AI, rather than creating intelligence, simply accelerates the existing complexity and inconsistencies of a broken system. When customer, policy, and billing data are spread across multiple platforms structured around products rather than people, it becomes nearly impossible to create a single, reliable view of the customer. I have seen organizations where establishing whether a single person holds two different policies is a major hurdle, which makes building a trusted data foundation for AI incredibly difficult. If the data feeding an AI model is incomplete or poorly governed, the resulting decisions will be difficult to justify and even harder to trust. You cannot layer modern agentic AI on top of legacy processes and manual workarounds and expect it to produce consistent, explainable results; it is like trying to build a high-speed rail system on top of a crumbling dirt road.
You’ve outlined five pillars of trust engineering—data, explainability, auditability, human oversight, and adaptability. Which of these do you believe is the most challenging for insurers to implement at scale?
While all five are essential, auditability is often the most difficult to implement because it requires a fundamental change in how insurance systems record their own history. In an AI-driven environment, you need to be able to reconstruct a decision months or years after the fact, capturing every rule, data point, and model output that existed at that precise millisecond. This isn’t just about defensive compliance; it’s about creating a basis for continuous learning and showing that the business is operating responsibly even as it scales. Most legacy systems were never designed for this level of forensic detail, and building that “audit trail” into a complex, multi-platform environment is a massive engineering undertaking. Without it, however, human oversight becomes a vague principle rather than a functional reality, leaving the company vulnerable to regulatory scrutiny and internal risk.
There is a common perception that strict governance and regulation act as a “drag” on innovation. How can robust trust engineering actually create a commercial advantage for an insurance firm?
The mistake is viewing governance as a brake on progress, when in reality, it is more like the brakes on a high-performance car—they allow you to go faster because you know you can stop or pivot safely. When trust engineering is built into the operating model, insurers can deploy new AI capabilities with much greater confidence, reducing the risk of a “stalled” innovation project due to last-minute compliance fears. It allows for a more sophisticated use of AI over time because the “control plane” for those models is already established and understood by the business. Furthermore, it streamlines the response to regulatory inquiries and customer complaints, turning what used to be a weeks-long manual investigation into a data-driven process that takes minutes. Ultimately, the insurers who can operationalize AI safely and transparently will be able to scale their innovations far faster than those who are constantly struggling to explain their own automated decisions.
In the move toward production-scale transformation, why is modern core architecture more critical than the specific AI models an insurer chooses to use?
The specific model is often the least important part of the equation because AI cannot function as a separate layer that sits apart from the core insurance business. If AI is influencing underwriting, pricing, or risk management, it must be deeply connected to the core systems where those decisions are actually executed and recorded. A modern core architecture brings data, business rules, and customer interactions into a single, cohesive flow, allowing for automation without a loss of visibility. Without this integration, there is a dangerous gap between the “intelligence” of the AI and the “action” of the policy lifecycle, leading to a lack of accountability. Moving from a successful pilot to enterprise-wide adoption requires a disciplined architecture that treats governance as a part of the workflow, not an afterthought.
What is your forecast for 2026?
By 2026, we will see a sharp divide in the market between the “leaders” who have mastered trust engineering and the “laggards” who are still struggling with opaque, fragmented systems. My forecast is that the industry’s focus will shift entirely from the novelty of AI models to the hard operational capability of “governed intelligence.” We will see the first major wave of regulatory enforcement under the EU AI Act, which will serve as a wake-up call for companies that treated AI as a side project rather than a core architectural shift. The most successful insurers will be those who have moved past the “pilot” phase to a production-scale model where every automated decision—from a health insurance premium to a complex commercial claim—is fully traceable, explainable, and adaptable to changing market conditions. In this environment, trust will no longer be a soft marketing term; it will be the primary metric of operational excellence and the ultimate driver of competitive growth.
