Only 50–60% of companies consistently use AI, and the rest are holding back because they do not trust it. The problem is not technical capability but a lack of guarantees: if an AI system makes a costly mistake, who pays? This project, co-created with AXA, builds an insurance model for AI. It will develop an assurance framework to audit AI systems for accuracy, bias, fairness, and privacy. It will then estimate the likelihood and financial size of claims when AI underperforms, and create algorithms that adjust insurance premiums dynamically as risk profiles change. If successful, insurance-backed performance guarantees will let companies deploy AI with confidence, knowing they can claim compensation if it fails. That could unlock the estimated $13 trillion in economic value AI promises, equivalent to the UK’s entire GDP. The work also gives regulators a scalable, market-based mechanism to manage AI risk—without needing to inspect every algorithm themselves.
View original technical description
AI is rapidly reshaping the way we work and live. Tailoring state-of-the-art solutions was only possible for those companies with AI research labs or individuals with deep technical knowledge of how the technology works. However, there is a growing commoditisation of AI; for example, companies such as OpenAI have provided affordable mass access to Large Language Models, with ChatGPT currently having around 180.5 million users. This increased access to AI could, McKinsey estimates, deliver an additional economic output of approximately US$13 trillion by 2030, with generative AI contributing $3.5 trillion. This is equivalent to the GDP of the UK. Despite the clear advantages, corporate adoption has plateaued, with only 50-60% of companies utilising AI technologies consistently. This stagnation is largely due to a prevalent distrust of AI among leaders and the general public, compounded by a limited understanding of the technology. The key objective of this project is to work with one of the world’s largest insurance companies to undertake world-leading research to provide insurance services that indemnify organisations against damage caused by underperforming or unreliable AI solutions. Reduced insurance premiums create incentives to develop and deploy high-quality auditable AI systems that inspire trust. Clear understanding and management of risks allows organisations to innovate with confidence. This will ultimately lead to better risk management practices and standards for AI. With insurance-backed performance guarantees, companies creating and using AI can offer compensation in case of underperformance, thus encouraging broader adoption of innovative AI solutions. Moreover, the AI insurance market can complement policymakers and regulators in managing AI risks by providing scalable evaluation mechanisms through the financial markets. We have worked with AXA to co-create a transformational research programme that addressed the key interconnected challenges that need to be addressed to create a robust AI insurance model: Risk Assessment and Measurement: This involves evaluating the potential risks associated with AI systems. For AI insurance, this means understanding how AI algorithms perform under various conditions and scenarios and involves quantifying the likelihood of failure or underperformance of algorithmic pipelines. This will require the development of an implementable AI assurance framework to allow robust audit of algorithmic systems across appropriate metrics including model accuracy, bias, fairness, and privacy. Claims Likelihood and Size Estimation: Once risks are measured, insurers need to estimate the potential size of claims and the likelihood of the risk being realised. This involves calculating the possible financial losses that clients might incur if the AI system fails to perform as expected. This estimation is crucial for setting premiums and reserves. This requires precise evaluation of the likelihood and severity of algorithmic systems’ failure across all the key evaluation metrics using scenarios and statistical modelling. Dynamic Insurance Premium Adjustment: Insurance premiums should adjust to the evolving risk profile of AI systems and corresponding claims estimation. As new data emerges and AI systems update, the risk profile changes, necessitating premium adjustments. This strategy ensures insurance products are priced accurately to reflect the true risk while staying competitive in the market. This will require developing algorithms that seamlessly assimilate available data while maintaining appropriate transparency and regulatory compliance.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know