News & Events

Artificial Intelligence and Insurance Coverage, Part 1: New AI Exclusions and Emerging Liability Issues

Key Takeaways

• Standard general liability policies may still provide coverage for some AI-related claims.
• New ISO generative AI exclusions suggest insurers view existing policies as potentially covering AI-related liabilities.
• The scope of AI exclusions may depend in part on how courts apply phrases like “arising out of.”
• Policyholders should carefully review renewal policies for new AI exclusions and negotiate where possible.
• AI-related risks may lead to new insurance products, similar to specialized cyber and environmental liability policies.

Artificial Intelligence (“AI”), especially so-called “generative AI,” is one of the hottest corporate and financial topics today.  Unsurprisingly, with new ways of working come new risks, and with new risks come new insurance coverage issues.  In this series of posts, we consider ideas from senior attorneys at Anderson & Kreiger on the role AI will play or already does play in the industries we represent and the insurance issues that may arise out of the use of AI.  In the first post, we take a look at the insurance industry’s first response to some of the new exposures created by the use of AI. Future posts will discuss these and other issues relating to the intersection between AI and liability insurance.

What Are the New ISO Generative AI Exclusions?

The Insurance Services Office (“ISO”), which, among other things, develops and publishes standard policy language often used by general liability and other insurers, has released its first set of exclusions for “Generative Artificial Intelligence.”  Different versions of the exclusion are written to apply to general liability policies as a whole (Form CG 40 47 01 26), only to “personal and advertising injury” coverage in general liability policies (Coverage B of standard general liability policies) (Form CG 40 48 01 26), and to products/complete operations liability coverage (CG 35 08 01 26).

These new general liability policy exclusions share a common core: they purport to exclude coverage for liability “arising out of ‘generative artificial intelligence,’” which is defined as “a machine-based learning system or model that is trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code.”

The proposed language raises several issues.  First, the existence of the exclusion is an implicit admission that nothing in the standard-form general liability policy excludes coverage for AI-related losses.  When faced with the argument that liability for AI-related losses was never contemplated by an insurer, the later creation and addition of this endorsement—either to the policyholder’s coverage or by insurers generally to their policies—are strong evidence that the industry generally and this insurer specifically recognized the potential that the unamended policy covered AI-related losses.

Second, the “arising out of” language will need to be carefully considered.  Most general liability policies include coverage for “personal and advertising injury,” which typically includes coverage for things like libel, disparagement, violations of rights of privacy, and copyright and trade dress infringement in an advertisement.  With respect to these sorts of offenses, one can imagine several types of claims, like copyright infringement or infringement of rights of publicity, where an AI-generated image, audio, video, or similar item will be the immediate and direct cause of the claimant’s alleged damages.

However, for other claims the events leading to the claim involve the use of AI, that use may stand at several removes from the ultimate claim or harm.  For example, consider the use of AI in construction and design or coding.  Generative AI could be incorporated into one of thousands of calculations or lines of code.  When a claim arises, is the claim barred regardless of the role of AI-generated work?  Is “but for” causation sufficient?  Must an insurer show that the AI-generated elements were the proximate cause of the loss?

Third, the definition of “generative artificial intelligence” in the new ISO exclusion is relatively narrow and should not be taken to apply to any and all AI.  The definition requires a “machine-based learning system or model.”  That system or model also must be “trained on data” and have “the ability to create content or responses.”  The exclusion appears aimed at the new generative AI tools, like Chat GPT, Claude, and Copilot, that can be used to create new images, content, or other work, and which use swaths of publicly available information (so it is claimed) to provide those models with data and feedback.  It does not appear to be aimed at so-called “traditional AI,” which does not “learn” or have the “ability to create” new work in the same way as generative AI and that is typically limited to algorithmic rules, tasks, and outputs based on a narrowly defined set of data or circumstances.  (Algorithmic and predictive search and recommendation tools, like those run by companies like Netflix and Facebook, are examples of traditional AI—those systems analyze data and make highly accurate predictions of behavior but do not create new content.)

The precise tools that the AI definition in the new exclusions intend to capture, and which, if any, it actually does capture, are bound to be the source of litigation as policyholders and insurers dispute the type of AI tool that was used and whether the “generative” features of that tool were the cause of any alleged liability.

Finally, there does not yet appear to be a standard exclusion in general use for other types of policies, like directors and officers insurance, errors and omissions coverage, and professional liability coverage.  However, there are indications that some insurers are beginning to add AI exclusions to these policies.  Even without express exclusions, those policies will raise their own set of issues.  Issues will likely arise as to whether and to what extent the use of AI falls within the scope of the insured risk.  For example, if a policy covers “professional services,” is AI just another tool used by a professional, or is work generated by AI outside the scope of the insured “professional services”?

On the flip side, new products may be developed to provide coverage for AI-related exposures.  Just as the insurance industry developed new products to cover environmental hazards and cyber/privacy risks, AI exposures may lead to a new line of insurance products.

What Should Policyholders Do About AI Exclusions?

As the insurance industry develops its responses to emerging AI-related exposures and claims, insureds should carefully review their policies as well as notices of new exclusions to identify any AI exclusions added to new or renewal policies and to consider whether their use of AI might lead to claims.  Insureds should also work with their brokers and advisors to assess whether the inclusion of an AI exclusion or its scope may be negotiable and to consider alternative coverage to fill any coverage gaps.