Few technologies have moved as quickly from emerging possibilities to strategic business issue as artificial intelligence. In early 2022, AI was already an important research field but for most organizations it remained an enabling technology: a tool for prediction, classification, recommendation, and automation. Four years later, that picture has changed considerably. AI is increasingly becoming an active participant in knowledge work, software development, and business processes. Over that period AI has undergone a fundamental shift – from systems designed primarily to respond to instructions toward systems capable of working through complex problems, coordinating actions and increasingly act on behalf of their users.
For WTW and the Willis Research Network (WRN), this evolution has meant that AI has moved from being principally a technology and analytics issue to a much broader question of risk, resilience, governance, and risk transfer. The speed of that progression is particularly revealing. In only a few years, questions about what AI might make possible have become questions about how organizations govern, insure and build resilience around systems they are already using.
AI becomes a broader risk question
In 2022, WTW was already examining the implications of AI for insurance. Its November 2022 analysis described how insurers were using AI and machine learning to automate workflows and improve loss prediction, while highlighting concerns about bias, transparency, explainability, and data controls. At the time, the central question was how insurers could capture AI's benefits without undermining trust in the decisions that increasingly affected customers.
From copilots to agents
The release of ChatGPT in late 2022 accelerated the conversation dramatically. Generative AI made advanced machine learning accessible to a mass audience and demonstrated that a single interface could perform tasks previously associated with specialized software: writing, summarizing, translating, coding and generating images. What looked in 2022 like a remarkable new interface has, by 2026, become the foundation for a much broader technology stack.
For WRN, this rapid development created a new research agenda. AI could no longer be considered simply as a tool that could improve insurance analytics; its rapidly changing capabilities raised questions across WTW’s lines of business. The WRN’s technology research program has expanded accordingly, now covering AI governance and liability, autonomous systems, technology trust, and the ways AI and data science are changing modelling and horizon scanning.
One of the most significant developments has been the transition from copilots to agents. WTW's own work has tracked this shift. By late 2025, WTW was describing agentic AI as the next frontier for insurance, as organizations began experimenting with systems able to undertake tasks with greater autonomy. This changes the nature of the risk: a system that recommends an action raises different questions of control and accountability from one that can initiate or execute it. The need for human validation, ethical controls and effective governance therefore becomes increasingly important.
Lower costs expand experimentation beyond the largest technology companies and make AI increasingly relevant to smaller businesses and specialized applications
AI has also become cheaper and more accessible. Improvements in chips, model architectures, inference, and smaller models have reduced the cost of deploying sophisticated AI. Lower costs expand experimentation beyond the largest technology companies and make AI increasingly relevant to smaller businesses and specialized applications.
At the same time, AI has also become increasingly multimodal. While the generative AI breakthrough of 2022 was dominated by text, by 2026, leading systems can work across text, images, audio, video and code, often within the same workflow. This expands AI from a tool for manipulating language into a more general interface for information and decision-making.
When AI moves into the physical world
That development connects AI to another disruptive technology trend: the convergence of software and physical systems. Robotics, autonomous vehicles, industrial automation and other forms of physical AI are becoming more capable as advances in perception, reasoning, and planning are incorporated into machines. As AI moves into the physical world, the consequences of its decisions can become physical too. Warehouses, factories, laboratories, transport systems and other physical environments can increasingly become domains in which AI makes decisions and coordinates actions.
This convergence also changes the risk landscape. AI is not simply software. It depends on computing power, data centers, electricity, telecommunications and advanced semiconductors. Its resilience is consequently connected to physical infrastructure and concentrated technology supply chains. The more deeply organizations embed AI in critical processes, the more a technology risk can become an operational or systemic risk
Understanding and insuring AI risk
As AI capabilities developed, WRN's research explored both the opportunities they created and the new risks they introduced. In 2024, WRN examined how generative AI could reshape scenario analysis in insurance. The research highlighted its potential to change how insurers develop scenarios used to understand tail risks, emerging risks, risk aggregation, and strategic uncertainty. AI was beginning to play a dual role: creating new exposures while also providing new ways to explore risks and scenarios that have not yet occurred.
As AI systems became more capable, questions of accountability became increasingly important
As AI systems became more capable, questions of accountability became increasingly important. Research with Drexel University's Anat Lior examined emerging practices around AI liability policies, asking how responsibility should be assigned when AI systems cause harm or fail to perform as expected. By 2026, WRN's work had developed further into questions of AI governance, trust, auditability, human impact and the practical application of responsible AI. Its June 2026 AI Governance in Practice panel brought together academic, legal, human-capital, and insurance perspectives to examine how governance works across real systems and evolving risk environments.
The same questions were emerging around insurance coverage. WTW's 2024 work highlighted how AI-related incidents could cut across cyber, professional liability, intellectual property, employment practices and directors and officers liability rather than fitting neatly within a single policy. By 2025, the discussion had moved toward "silent AI" – the possibility that AI-related losses could fall across existing insurance policies without being explicitly addressed. This made mapping AI exposures against existing coverage and identifying potential gaps increasingly important for organizations and insurers.
AI increasingly sits on both sides of the risk equation
At the same time, WTW has examined AI as an opportunity for the insurance industry itself. Its 2024 advanced analytics survey found that 49% of insurers were incorporating AI into their analytics processes. By 2026, WTW reported that insurers making greater investments in advanced analytics and AI were achieving measurable improvements in profitability and premium growth. AI increasingly sits on both sides of the risk equation: creating exposures that organisations need to understand and manage, while providing new tools to improve analytics and decision-making.
The shrinking gap between innovation and risk
For WRN and WTW, this evolution illustrates a broader lesson about disruptive technology. WRN's role is to investigate emerging questions before they become established risks, connecting academic research and forward-looking analysis to practical applications. WTW then brings those insights into risk advisory, analytics, insurance, and client decision-making. Between 2022 and 2026, AI demonstrates why that connection matters: the time between a technology emerging and its embedding in critical systems is becoming shorter.
The next challenge is keeping pace with how quickly new AI capabilities move from experimentation into the systems on which organizations depend. The experience since 2022 suggests that governance, resilience, risk management and insurance may have increasingly little time to catch up.
Authors
