Experts caution against blanket AI exclusions in management liability policies

(The Insurer) - Insurers should avoid sweeping artificial intelligence exclusions in management liability policies, experts say, warning that such measures risk undermining coverage and urging instead for narrowly defined language as AI use and regulations evolve.

“If you’re not specifically targeting certain uses of certain types of technology, it could be very easy to show how radical absolute exclusion is going to just eviscerate all coverage for everything,” Hunton Andrews Kurth partner Mike Levine told The Insurer.

The warning comes as WR Berkley recently drafted a broad AI exclusion for its directors and officers, errors and omissions and fiduciary liability products. The proposed language would bar coverage for “any actual or alleged use, deployment, or development of Artificial Intelligence.”

While not yet implemented commercially, Levine said the draft highlights the mounting challenge insurers face as they seek to limit AI-related exposures.

WR Berkley did not respond to a request to contribute to this article.

Levine noted that defining AI remains a key obstacle.

“The definition of artificial intelligence is incredibly broad and really without limitation,” he said. “When you start to think about how AI is being deployed in all of the different systems, it becomes very difficult to put boundaries around it.”

This ambiguity poses practical problems for insurers and policyholders alike.

“Exclusions have to be clear on their face and workable for claim handlers, adjusters, judges, and lawyers, not just computer scientists,” Levine explained.

He noted that the management liability sector is already experiencing increased AI-related claims, particularly in securities litigation involving alleged misstatements about AI capabilities, sometimes referred to as “AI washing.”

“We’ve seen lawsuits precipitating negative stock movements, leading to shareholder class actions,” Levine said. “The only thing new here is the statement concerned artificial intelligence and not something else.”

Despite the uptick in claims, insurers have hesitated to implement blanket exclusions because of competitive pressures.

“Nobody wants to be the first to roll out that broad exclusion,” Levine said. “As soon as you do that, you alienate a large segment of your policyholders, who will go to your competitors.”

The ubiquity of AI further complicates the issue, creating hidden risks for policyholders who may unknowingly lose coverage.

"If the goal is to exclude anything and everything that touches on or arises out of artificial intelligence, it's hard to point to something nowadays that doesn't," Levine said.

“You may be using something that’s utilizing AI and you don’t even know it,” Levine warned. “That becomes very dangerous, because as a policyholder, you want to know what’s covered and what’s not covered.”

Levine predicted that effective exclusions will need to be narrowly targeted, both in how AI is defined and in the specific risks excluded.

“The effective initial exclusions that we’re going to see will be focused and narrow,” he said, suggesting management liability exclusions might address misstatements about AI, rather than barring all AI-related activity.

He called for greater collaboration between insurers, brokers and policyholders to develop workable exclusion language. “There’s got to be a dialog to come up with language that makes sense and is functional,” Levine said. “It’s the lack of functionality that is the biggest problem.”

RT ProExec executive vice president John Delaplane echoed the need for a measured approach, noting the rapid advancement and diversity of AI technologies makes it difficult to encapsulate the full scope within a single definition.

“What could be considered advanced AI today could just become standard technology tomorrow,” he said. “The dynamic nature of AI in the near term is going to make it difficult to adopt any sort of AI absolute language, because, at the end of the day, that's just going to make it difficult to facilitate consistency and intent of coverage.”

Delaplane pointed to the insurance market’s history in handling emerging risks without blanket exclusions.

“If you look at the past five years with litigation surrounding the Me Too movement, cybersecurity, ESG and COVID-19, AI is not much different than those items. The marketplace has navigated these events and trends without implementing across-the-board exclusionary language for any of them,” he said.

On a macro level, Delaplane does not foresee the industry embracing absolute exclusions.

“I don't see us adopting it across the board, on an absolute basis. On the flip side, we're seeing the marketplace create AI-specific insurance products. So I think it just comes down to each insurance company's comfort, appetite and risk tolerance for AI moving forward.”

Looking ahead, Levine highlighted that companies will need to improve their understanding and governance of AI usage across operations.

"You're going to have to have a chief AI officer that can engage with all stakeholders and understand how that company is utilizing artificial intelligence in everything it does, from hiring to distribution to manufacturing to accounting, basically everything they do."

The broad Berkley exclusion “may never see the light of day commercially” due to its fundamental flaws, Levine predicted, noting that the industry will need to adopt more nuanced approaches to address AI-related risks without undermining the value of insurance coverage.