As AI use spreads through UK businesses, brokers are facing new questions before they can take a cyber risk to market.
Simon Hughes, chief commercial officer at Cowbell, the US-headquartered cyber insurer he helped launch in the UK, said AI has not changed the basic fundamentals of what makes a cyber risk difficult to place. What is changing is where weaknesses can emerge.
Alongside the malicious use of AI by third parties to produce more convincing phishing and social engineering, Hughes pointed to a less obvious threat: staff connecting AI tools to email, file stores and messaging platforms without anyone weighing up what those tools can access.
"Every single business in the country is almost certainly going to use some sort of AI, even now or in the future. And it's too easy to get sucked into that conversation and give away data, give away everything, quite frankly, that your business relies on," Hughes said.
The checks previously relied on have not changed because of AI – employee training and clean application forms remain the baseline, just as they were before.
"It's still the same fundamentals that brokers need to look out for, for a good risk or a bad risk. And a bad risk, I mean, is difficult to insure," Hughes said.
That leaves room to help clients stay ahead of AI-driven cyber risks by establishing how the technology is actually being used, particularly in regulated sectors handling sensitive information.
"You can have all this data at your fingertips, but at what cost? Especially if you haven't got an enterprise-level idea of how you want to safeguard it," Hughes said. "Who else can access that type of data?"
Cowbell has not yet seen significant problems emerging from AI exposure at renewal, though Hughes suspects it is a conversation few of its broker partners are raising proactively. In his view, what can materially change an insurer's assessment is evidence that the client has considered the exposure before something goes wrong.
"My viewpoint of an insured would be significantly changed if they came to us and said, you know what, we have connected [AI] to every single one of our terabytes of data to maximize our ability to analyze our data. Is that a problem or is that an issue? Them thinking about it in the first place rather than just relying on the insurance," Hughes said.
"Clearly, improvements around the guardrails they put on the use of AI is the most important factor when it comes to the non-nefarious use of AI."
Even strong corporate controls have limits. A business may invest in enterprise AI licences and impose guardrails on approved systems, but employees can still use personal devices and accounts to put company information into tools outside those controls.
"If I've got my own [AI] on my phone and I want to take a picture of something and say, can you do this on my screen, that's a massive loophole in that, and it's just beyond easy to be able to do that type of thing," Hughes said.
He cited healthcare as an example of how easily that could result in sensitive information being exposed. Businesses remain accountable to the Information Commissioner's Office (ICO) for resulting data breaches, regardless of which AI tool or device was involved – making it a blind spot clients need to have considered before renewal.
For SMEs without large cyber budgets, that consideration does not have to mean expensive controls. Hughes said staff training remains one of the most valuable options available, giving businesses a relatively simple way to reduce the risk of employees making the wrong decision – a discipline that sits alongside the more sophisticated ways brokers are already applying AI to sharpen their own underwriting input.
"The number one thing a limited budget company can do when it comes to their IT security is just get training for their staff. As basic and as low level as you think that might be, just that knowledge that, oh, that looks a bit weird, shall I double-check that or shall I click on it and see what happens? Just that pause, just that reflective moment, and just knowing to do it, and that will save a lot," he said.
There are circumstances where that conversation should happen before insurers are approached at all.
"I think brokers have a good eye for what's insurable and what's not insurable," Hughes said.
Early-stage businesses with few policies or controls in place and little evidence that cyber risk has been considered are one example. Some, he said, are effectively "shell companies" built around AI plugins, with little security infrastructure of their own and an assumption that cloud or AI providers will take care of data security.
Those cases remain the exception. Hughes said insurers can accommodate significant AI exposure, including both accidental misuse and malicious third-party activity.
That puts the emphasis less on whether a client uses AI than on what happens before renewal. A broker who can establish what tools are being used, what data they can reach, who controls that access and what happens when employees bypass those controls can give an underwriter something much more useful than an assurance that the business simply has an AI policy.
The difference between a difficult cyber risk and a placeable one may increasingly be evidence that somebody has asked those questions before the insurer does.
