Consumers are increasingly turning to artificial intelligence for answers about insurance before they ever speak to a broker. At the same time, brokers are using AI to compare policy wordings, draft client communications and process information more efficiently, while businesses are embedding the technology into everyday operations.
As AI moves closer to the customer, however, one question remains largely unanswered: what happens when it gives the wrong insurance advice?
As regulators and the insurance industry examine how AI could reshape financial services, the question of liability is becoming harder to ignore. The debate is becoming less about whether firms should adopt AI and more about who carries responsibility when automated recommendations influence decisions with financial consequences.
For Peter Robinson, director at Prizm Solutions, AI is already an everyday business tool rather than an emerging technology. His firm uses it to compare policy wordings, upload information and assist with drafting sales pitches, freeing brokers to focus on client work.
"I honestly think brokers who don't adopt AI will struggle to continue to compete," he said.
But Robinson believes the bigger shift could come if customers begin relying on AI for advice before approaching a broker at all.
"Long term I do worry if brokers will become semi-redundant because business owners and consumers will simply type into their AI of choice what they want and it will 'advise' them accordingly."
Robinson believes AI's tendency to generate confident but incorrect answers raises what he sees as the industry's biggest unresolved question.
"My understanding is that AI contains a feature called catastrophic failure which means if it doesn't know the answer, it will make it up and if a client relies on AI for advice and it's wrong, will Mr Google or Mr Grok be liable?"
That question extends beyond brokers. As businesses adopt AI more widely, insurers are also grappling with where liability ultimately sits if automated systems generate inaccurate advice or recommendations.
"I think the gaps... who's liable? Is it the programmer? Is it the client? Is it the user? Is it the person who's selling it? That's not clear," said Ed Ventham, head of broking at Assured Cyber.
Ventham believes existing cyber insurance can already respond to many AI-related exposures because AI systems are ultimately computer systems. The more difficult questions concern risks that do not fit neatly within traditional policy structures.
"Cyber liability can pick up a lot of the exposures that AI brings... but there's bits that fall outside of it," he said.
Hallucinations, poor model performance and situations where AI systems behave in unexpected ways do not fit neatly within existing cyber or professional indemnity products, he said. That leaves many organisations adopting AI before they fully understand where their existing insurance may no longer respond.
"Everyone's now using it. They're just working out what the risk is."
While liability questions remain unresolved, the Chartered Insurance Institute (CII) believes firms can mitigate many of the risks through strong governance. The professional body said AI has the potential to improve customer outcomes through more accurate risk assessment and wider access to financial services. However, it said those benefits will depend on firms maintaining strong governance, ethical oversight and human judgement as AI becomes more influential in decision-making.
The CII said AI should enhance, rather than replace, professional expertise, with people remaining responsible for customer focus, ethical decision-making and validating automated outputs where appropriate.
Robinson believes that leaves brokers with an enduring role, particularly where insurance needs become more complex.
"I am convinced brokers can continue to add value especially for more complicated requirements because we know the marketplace and can tailor the insurance protection to the clients' needs and requirements."
For Ventham, however, the industry's biggest questions are unlikely to be answered until AI-generated advice is tested in the real world.
"We need to see some claims to work out how this is actually going to evolve."
