They’ve arrived. Be prepared.
AI controls the operator as much as the operator controls the AI,
but only the user and/or their employer can be liable for damage or injury to others.
By Joseph S. Harrington, CPCU
Imagine, if you can, an insurance policy written with the following exclusion:
“The Insurer shall not be liable [for] . . . any actual or alleged use, deployment, or development of a computer or computers by any person or entity . . .
You would undoubtedly scoff at such a provision. Just about everything today is done using computers in one way or another. Excluding coverage for computer-related losses would effectively render the policy useless.
Yet the citation above draws upon language a recent exclusion developed by Berkley, a leading London-based intermediary, except that in place of “computers,” the provision addresses “artificial intelligence.”
While insurers are rapidly integrating “AI” into their own operations, there are only just beginning to address their level of exposure to claims resulting from AI. Among other things, AI providers have been sued for defamation over inaccurate and disparaging information reported by AI systems.
In response, several major insurance providers—including AIG, Berkley, Chubb, Great American, and QBE—have developed endorsements or policy provisions restricting coverage for AI claims. The exclusion referred to above is for use in Berkley’s management and professional liability products, but there’s consideration among insurers to extend such exclusions to other lines.
An ‘absolute’ approach
Berkley’s initiative, characterized as an “absolute” AI exclusion, is long and comprehensive, eliminating coverage for an extensive list of actual or alleged losses. In paraphrase, the exclusion applies to:
- The generation, creation, or dissemination of content or communications using AI, including but not limited to:
- Failure to detect third party AI content or communications;
- Inadequate policies, practices, procedures, or training related to AI;
- Breaches of obligations regarding the development, deployment, identification, or containment of AI;
- products or services sold, distributed, performed, or utilized that incorporates AI; or
- representations, warranties, promises, or agreements made by a chatbot or virtual customer service agent.
- Statements, disclosures, or representations related to AI, including but not limited to:
- The use, deployment, development, or integration of AI into insured operations;
- Assessments of the insured’s risks or vulnerabilities arising from AI;
- Business plans, capabilities, or opportunities involving AI;
- Violations of any laws or regulations regarding the development, use, and disclosure of AI; or
- Demands, requests, or orders that the insured investigate or respond to the risks, or impacts of AI.
The firm’s comprehensive AI exclusion is buttressed by a comprehensive definition of “artificial intelligence,” which states in part that AI consists of–
“any machine-based system that . . . infers, from input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments . . .”
The definition also includes systems that “emulate the structure and characteristics of input data in order to generate derived synthetic content,” such as text, images, audio, and other digital content.
So what’s covered?
By now, virtually everyone has already encountered AI. Online inquiries are now routinely routed and/or responded to by AI. AI instantly generates reports in response to search engine inquiries. AI knows what you buy and what you want to buy, sometimes before you do.
Moreover, AI is being built into computer applications of all kinds, to the point where it’s practically impossible to avoid using AI in some capacity. As AI becomes ubiquitous, an “absolute” AI exclusion would void just about all coverage under a policy, much like our hypothetical “computer” exclusion mentioned at the outset.
The real impact of AI exclusions will likely be how they set the table for limited coverage provided for a specified scope of AI activity. That seems to be the case for Chubb, which covers some AI-related losses but excludes coverage for incidents resulted in “widespread” losses simultaneously. For its part, QBE has introduced a separate limit, up to 2.5% of the insured limit, covering fines assessed under the European Union’s Artificial Intelligence Act.
‘Silent’ coverage question
Pairing “absolute” AI exclusions with targeted coverage provisions seeks to do for artificial intelligence what insurers have sought to do with “cyber” insurance: eliminate “silent” coverage as an exposure insured under the basic policy and replace it with “affirmative” coverage provided on a narrower basis.
We can see a progression of exposure over time.
Early on, computers dramatically changed how work was done, but they were self-contained and carried out instructions input by their operators, with little or no effect on the liability of the users.
“Cyber” coverage arose to address the potential for losses due to malfunctions or malicious actors outside the control of the insured. Efforts to eliminate “silent” cyber coverage are essentially efforts to segregate cyber perils from traditional causes of loss.
Whether that can really be done remains a question, however. A cyber event may be the proximate cause of a loss, but that event could also result from negligence or misjudgment subject to management, professional, or general liability. A cyber event can also contribute to a first-party property loss.
The next level
Artificial intelligence takes this exposure to a whole new level.
AI makes decisions, and refines its own decision-making processes, in ways an insured may not fully understand. AI controls the operator as much as the operator controls the AI, but only the user and/or their employer can be liable for damage or injury to others.
AI is also connected to global networks, creating the potential for a user to be the instigator or victim of a massive disruption in computer operations.
For 2026 and beyond, agents, brokers, and their clients are going to have to understand where and how clients are utilizing AI, and whether they are insured for it.
The author
Joseph S. Harrington, CPCU, is an independent business writer specializing in property and casualty insurance coverages and operations. For 21 years, Joe was the communications director for the American Association of Insurance Services (AAIS), a P&C advisory organization. Prior to that, Joe worked in journalism and as a reporter and editor in financial services
