Artificial intelligence is becoming a foundational layer of how modern business operates. From product development to risk assessment, AI is increasingly empowering decision-making in multiple directions. Yet with this rapid adoption comes a growing set of challenges, particularly around how we define, govern, and insure against AI-driven risks. What happens when AI gets it wrong?
Insurance has a critical role to play, not just as a backstop for failure but as a force that enables innovation to move forward responsibly and confidently.
But as we embrace AI’s potential, we must also confront its complexity. Many of the risks it introduces are latent and not yet fully understood – what some are calling “silent AI”. These exposures are already embedded within current systems and processes, often without being explicitly recognised or priced into insurance coverage. This mirrors the industry’s earlier experience with “silent cyber”, where risk was assumed but not always acknowledged until a major loss forced a reassessment.
One of the central challenges lies in attribution. AI models don’t behave like static tools. They evolve, adapt, and sometimes act unpredictably. The same model can produce different outcomes without a single change in its underlying code. This makes assigning liability difficult. In most scenarios, responsibility will likely be shared between the developer of the model and the business deploying it. This concept of joint liability must be better understood if we are to develop sustainable, insurable frameworks.
Compounding the issue is a significant educational gap on all sides. Too often, AI is treated as a black box. A more accessible definition, as any system that performs automated decision-making, can help demystify its reach and invite more people into the conversation. Only by understanding how these systems function can insurers, regulators, and businesses engage meaningfully in conversations.
The pace at which AI is advancing presents an additional layer of complexity. Regulation can't keep up. In this vacuum, some organisations are rushing ahead without governance; others are paralysed by uncertainty. Both responses are risky. The key lies in supporting structured, responsible experimentation, where innovation is encouraged, but within a framework that prioritises transparency, safety, and resilience.
To navigate this terrain, the insurance sector must equip itself with new tools and mindsets. That means moving beyond passive risk assessment and becoming an active partner in AI adoption. It means designing products that address AI-related liability explicitly rather than waiting for loss events to expose the gaps; driving a deeper, more responsive understanding of how AI is already shaping portfolio risk.
At Carbon, this philosophy is already being put into practice. Our in-house platform, Graphene, is built from the ground up on a modern tech stack to challenge traditional assumptions around data ingestion, analysis, and insight-sharing. By standardising and augmenting data, and layering on machine learning and other models, Graphene equips underwriters with real-time, AI-informed insights across binders, Coverholders, territories, and lines of business. Crucially, those insights are shared with all stakeholders, democratising access to intelligence and ensuring that decisions are made with the full picture in view.
In this way, Graphene is a real-world example of how AI and insurance can co-evolve. It shows that with the right infrastructure and mindset, insurers can drive innovation, transforming how we see risk and how we rise to meet it.
GET IN TOUCH: If you're interested in finding out more about Carbon, or would like a demo of our data platform Graphene, please reach out at info@carbonuw.com
