A leading expert has warned insurers are reconsidering their approach to the coverage of artificial intelligence risks amid the possibility that systemic threats may well be closer than many imagine.

Matthew Geyman, managing director, Intersys told Emerging Risks concerns were mounting and with it came a revision in insurers’ approach to the threats.

“Reports suggest that insurers across the market are withdrawing or sharply restricting coverage for artificial intelligence-related liabilities,” he explained. “This shift marks a deeper inflection point in how technological innovation is assessed and underwritten.

“Several major carriers are now publicly seeking to exclude AI exposures from their policies, signalling that today’s risk models are not yet equipped for the behaviour of increasingly autonomous systems. At the centre of this retreat is a structural challenge: AI introduces systemic, highly correlated tail-risk that traditional insurance mechanisms were never designed to absorb.

“Insurance operates effectively when losses are independent rather than systemic. A defective product or a professional error at one business does not trigger simultaneous failures across many others. AI disrupts this core assumption.”

Geyman explained: “A widely deployed model that malfunctions due to flawed training data, a software update or an unseen design vulnerability could generate claims across multiple sectors at the same moment. This creates the kind of aggregation risk that undermines the diversification insurers rely on to stabilise portfolios. It is not simply the magnitude of a single payout that concerns insurers, but the potential for concentrated spikes of loss that unfold in parallel.

“Recent legal developments indicate that AI’s systemic risk profile may be closer than many organisations assume. In November, the Munich Regional Court ruled that OpenAI was liable for the unlicensed use of copyrighted material in both the training and output of its models, rejecting the argument that text-and-data-mining exceptions applied where protected works were reproduced. Crucially, the ruling challenges the long-held assumption that copyright liability rests solely with model originators.

“If that reasoning is extended, the implications for insurers and insureds are profound. Software, content, analytics or decision-making tools developed using generative AI could embed latent copyright infringement at scale. In such a scenario, liability would not arise from isolated misuse, but from millions of ordinary, good-faith deployments by end users who may have no visibility into the provenance of training data.”

He cautioned: “This is the definition of systemic exposure. A single judicial interpretation could trigger coordinated claims across sectors, jurisdictions and balance sheets simultaneously. For insurers, the risk is not merely one large loss, but a cascading aggregation event driven by shared dependency on the same underlying models and datasets.

“From an underwriting perspective, this raises uncomfortable questions. If AI-assisted products create infringement risk that cannot be cleanly ring-fenced between developers and users, then traditional liability allocation breaks down. Exclusions, indemnities and vendor warranties begin to look less like risk transfer and more like attempts to contain an exposure that has already become structural.”

However, Geyman says as with any risk challenges will create opportunities.

“Within this challenge lies an opportunity for insurers to become, inadvertently, champions of AI safety,” he explained. “Insurers naturally incentivise risk reduction. They price more favourably when uncertainty is lowered and behaviours more predictable.

“By requiring clear documentation of model development, evidence of testing and validation, continuous monitoring and credible incident response plans, insurers can help raise the baseline of AI governance across the economy. Organisations that can demonstrate accountable and transparent AI lifecycles will ultimately secure more favourable terms. In this way, underwriting discipline may become a catalyst for safer and more responsible AI deployment.”

Geyman added: “Market innovation is already beginning. Certain carriers are experimenting with specialist AI liability products that link coverage to specific controls or independent audits. These resemble early cyber insurance models: initially narrow, data-scarce and expensive, but maturing as better risk information emerges. They represent a pragmatic bridge between the need to protect businesses and the lack of actuarial history needed to price systemic failures.”

Looking to the year ahead Geyman said three trends will shape how AI risk is managed and insured.

“First, AI insurance will evolve into a specialist class, with coverage contingent on demonstrable governance and continuous validation,” he explained. “Second, organisations will expand their internal AI risk engineering capabilities as they retain more liability.

“Third, regulatory and underwriting standards will converge, creating shared expectations that make AI behaviour more predictable and ultimately more insurable.

“The current caution among insurers is not a rejection of AI, but an invitation to modernise the frameworks through which risk is assessed and mitigated, and to avoid AI becoming the ultimate systemic risk. If insurers, regulators and technology developers can align around robust, measurable standards, the result will be an AI ecosystem that is safer, more transparent and more resilient: one in which innovation can advance with greater confidence and accountability.”