Insurers risk ‘silent AI’ without exposure visibility: BCG

Insurers can start covering emerging AI risks before a mature claims history exists, but only if they can identify and control accumulating exposures, a BCG managing director and partner tells Monte Carlo Today. 

Key points:
US tech investment could top $800bn in 2026
$150–200bn a year going into AI deployment
Data centres create new cover demand

Insurers can start covering emerging AI risks before they have years of claims data, but they need to understand where exposures are accumulating before they broaden that cover, according to Boston Consulting Group.

Raphael Troitzsch, a managing director and partner at BCG, said the speed at which AI risks were developing meant insurers could not rely solely on the historical experience normally used to price emerging risks.

“There is a genuine data problem and a genuine speed problem,” he told Monte Carlo Today.

“Accumulation is dangerous if you’re not aware of it. It’s manageable if you can surface it.”

Rather than waiting for a mature claims history, Troitzsch said carriers could begin with clearly bounded risks, specific perils and defined limits, then widen coverage as experience and data accumulated.

But doing so required insurers to understand the AI exposures already sitting across their portfolios – and the potential for a single event to generate losses across multiple lines.

“Accumulation is dangerous if you’re not aware of it. It’s manageable if you can surface it,” he said.

Troitzsch drew a parallel with silent cyber, where insurers discovered after losses that traditional policies could expose them to cyber risks they had neither explicitly identified nor priced.

“Let’s not repeat that silent cyber treatment,” he said.

‘Gap hiding in plain sight’

The opportunity is growing rapidly. Companies are investing heavily both in the physical infrastructure supporting AI and in deploying the technology across their operations, creating potential new property, liability and business interruption exposures.

BCG estimates that capital expenditure by the largest US technology companies exceeded $400 billion in 2025 and could surpass $800 billion in 2026, with spending increasingly shifting towards the data-centre infrastructure required to support AI.

“The insurance industry alone, I think, is going to struggle to cover all of that,” Troitzsch said. “But for sure, that’s a growth opportunity.”

He also pointed to companies investing $150 billion to $200 billion annually in deploying AI within their own operations, describing the resulting insurance need as a “gap hiding in plain sight”.

Liability is among the emerging questions. Traditional liability cover has largely been built around human decision-making, Troitzsch said. As AI models increasingly inform decisions and actions, insurers will need to determine where responsibility sits between developers, deployers and users.

Business interruption presents another challenge. Troitzsch noted that policies typically require a physical-damage trigger and drew a parallel with the questions insurers faced over non-damage business interruption during Covid.

“At the moment, some of those outages that result aren’t covered at all,” he said.

Finding the exposure

The challenge is not only deciding which new risks insurers are prepared to cover. They also need to identify AI-related exposures that may already exist within conventional portfolios.

A failure involving a widely used model or AI provider could potentially generate losses across multiple insureds and lines of business.  Troitzsch stressed that makes portfolio visibility central to determining how far insurers can expand coverage.

AI itself could help solve part of the visibility problem. Troitzsch said the technology could be used to analyse portfolios and surface patterns and concentrations that might otherwise be difficult to identify.

But greater use of AI within insurance should not mean handing ultimate decision-making to autonomous systems.

“The human must remain in control,” he said. “Advising clients, interpreting nuanced risks, and weighing trade-offs. AI agents can do routine work, like reading information, cataloguing, summarising and surfacing patterns.”

Troitzsch said the aim is to use the technology to strengthen rather than replace human judgement. “This is really about teaching the machine enough so that it can help you and supercharge you as a human.”

No time to wait

Ahead of Monte Carlo, Troitzsch said he had encountered a view in recent months that “the market will fix it” and insurers could afford to wait and see.

He disagreed. The industry needed to tackle the emerging risks directly if it wanted to capture the opportunity accompanying AI investment, he noted.

“It’s going to be an opportunity only for those carriers who are going to manage solving the visibility and coverage architecture problem first,” he said.

Raphael Troitzsch is a managing director and partner at BCG. He can be reached at: Troitzsch.Raphael@bcg.com

For more news from Monte Carlo Today, click here.

Did you get value from this story? Sign up to our free daily newsletters and get stories like this sent straight to your inbox.