AI risks measurable, often insurable: Experts
Reprints
SAN DIEGO – Emerging risks associated with corporate use of artificial intelligence can be quantified and transferred to insurers via existing policies, new AI policies and captives, a panel of experts said.
Generative AI is being increasingly used by various organizations for tasks such as customer service, said Michael Berger, Palo Alto, California-based head of insure AI at Munich Reinsurance Co.
One common vulnerability of the tools is the risk of AI “hallucinations” where they present false or misleading information as true, he said during a session Wednesday at Riskworld, the Risk & Insurance Management Society Inc.’s annual conference.
To quantify the risk, a data set showing answers provided by AI tools to questions can be analyzed to determine the hallucination rate, Mr. Berger said.
“If we are utilizing similar models for similar use cases, then the error rates of those models can be correlated,” he said.
Once the risks are quantified, they can be transferred, often through existing policies, Mr. Berger said. These policies range from property policies, where an AI failure may lead to property damage, to technology errors and omissions policies.
In addition, specialty AI insurance policies are being developed by companies, including Munich Re, to cover risks that would not be covered by traditional insurance, he said.
Companies may also consider using captives to cover AI risks, said Joe Rosenberger, chief captive analyst at the North Carolina Department of Insurance in Raleigh.
“Since captive insurance is self-insurance, you’re able to really personalize the policies,” he said.
In addition, captives can issue difference in conditions policies to cover AI risks, Mr. Rosenberger said.
