AI Agents Expose Insurance Blind Spots, Report Says
More than 90% of insurers’ exposure to AI agents may sit in policies that neither clearly cover nor exclude the technology.
News
Image Credit- Chetan Jha/ MIT Sloan Management Review India
Insurers may already be covering AI-related losses through conventional policies that were never designed or priced for the technology, according to a new report.
More than 90% of insurers’ exposure to risks from AI agents was contained in such “silent” cover as of March, the report said. The exposure was concentrated in cyber, directors and officers, commercial general liability and technology errors and omissions policies.
Silent cover refers to risks that are neither expressly covered nor excluded, potentially leaving insurers liable for losses they did not price into a policy.
The report, ‘Underwriting the Agent Economy,’ was led by researchers at the Artificial Intelligence Underwriting Company, or AIUC, a San Francisco-based business that certifies and insures AI agents. Researchers from Anthropic and OpenAI were among its co-authors, alongside experts from insurers, brokers, universities and research groups.
The report said insurers’ exposure could grow as businesses adopt AI agents that can operate software, access company data, move funds and take other actions with limited or no human supervision.
Unlike chatbots that mainly generate answers, AI agents can carry out tasks. Their failures could lead to claims involving professional negligence, data leaks, fraud, discrimination, cyberattacks or physical injury. Such losses may fall under several kinds of insurance, leading to disputes over which policy applies and whether the insurer must pay, the report said.
Recent cases show how those questions are beginning to arise.
US solar installer Wolf River Electric has sued Google after its AI Overviews feature allegedly published false claims that the company was facing legal action over deceptive business practices.
Wolf River is seeking at least $110 million in damages, alleging that the false information caused customers to cancel contracts. Google is contesting the lawsuit.
In Canada, Air Canada was ordered to compensate a passenger who relied on incorrect information supplied by the airline’s customer-service chatbot about bereavement fares.
Air Canada argued that it should not be held liable for information supplied by the chatbot. The British Columbia Civil Resolution Tribunal rejected that position, ruling that the airline was responsible for information presented on its website.
The report also cited the growing use of AI in fraud.
British engineering group Arup lost HK$200 million, about $25 million at the time, in 2024 after criminals used digitally generated versions of senior employees during a video call. An employee in Hong Kong was persuaded to transfer money to accounts controlled by the fraudsters.
A claim arising from such an incident could involve crime or cyber insurance, as well as social-engineering cover. An insurer might argue that the employee voluntarily approved the transfers, while the company could contend that the authorization was obtained through deception.
Professional liability claims could prove equally difficult to assign.
A company using an AI agent may blame the system’s developer for an error. The developer may argue that the customer configured the agent incorrectly, gave it excessive access or failed to supervise it.
The report said existing policies often do not resolve those questions because their terms do not mention AI, even though they may cover the resulting harm. It described that ambiguity as a legal dispute between insurers and policyholders waiting to happen.
A severe AI event could cause about $100 billion in direct damage, according to a scenario in the report. The wider economic cost could reach trillions of dollars if insurers withdrew cover, businesses slowed AI adoption, and investors pulled back from companies dependent on the technology.
The figure is a risk scenario, not a forecast.
The report warned that a failure involving a widely used model or service could affect hundreds or thousands of companies at once.
Kevin Kalinich, head of intangible assets at insurance broker Aon and a co-author of the report, said AI could produce “aggregated, systemic, correlated” losses.
The authors compared the risk with the disruption to terrorism insurance after the September 11 attacks. More than $40 billion in insured losses led carriers to restrict cover until governments introduced financial backstops, the report said.
Some insurance executives and brokers have argued that the warnings are overstated and could help companies sell dedicated AI insurance products.
That criticism is relevant because AIUC certifies AI agents and helps arrange insurance against losses caused by them, giving it a commercial interest in the growth of specialist cover.
The report nevertheless found that conventional policies were already creating uncertainty for both insurers and businesses. Some carriers have begun introducing exclusions for generative AI, which could reduce insurers’ hidden exposure while leaving policyholders with less protection.
It called for dedicated AI cover, common technical standards, clearer policy wording and continuing audits of AI systems.
Insurers would need to assess what an AI agent can do, which systems it can access, how it is supervised and whether its actions can be stopped or reversed, the report said.
Without clearer terms, major AI losses are likely to produce disputes over whether conventional policies cover risks that insurers did not know they had accepted.
