Businesses deploying AI agents may have less opportunity to intervene when something goes wrong, creating an exposure that some insurers are considering underwriting and pricing differently
As AI systems interact with one another and accelerate transactions, the presence of a human supervisor can become a less meaningful safeguard. Brokers assessing those deployments face a more demanding conversation about how clients use the technology and where responsibility sits when it fails.
According to Ashley Bennett Jones (pictured), senior vice president, AI Risk and Advisory, within Lockton’s Cyber & Technology business, agentic AI presents a heightened risk landscape that could influence how carriers distinguish between deployments.
“Especially when you’re talking about agent-to-agent interactions, there is a rapid acceleration of transactions. There is less opportunity for the human in the loop to one, exist in a meaningful way, or two, mitigate risk,” Jones told Insurance Business, noting some carriers have identified those exposures as an area they may want to underwrite and charge for differently.
Underwriting questions about AI are already becoming more common, particularly across cyber and technology insurance. Some carriers are issuing supplemental applications that examine governance and the technology businesses use.
The questions extend to whether companies maintain an inventory of their AI tools and use cases, and who is responsible for each deployment. A broker preparing that submission needs information about how AI operates within the business, beyond a general acknowledgment that the company uses it.
“I would be very surprised if on every line of coverage a year from now, it wasn’t commonplace to have a whole part of the underwriting discussion be about artificial intelligence,” she said. “We are actually taking the questions that we hear in the underwriting conversations as well as the supplemental applications that we have seen in the marketplace and utilizing those to help build out the conversations with our clients in advance.”
Coverage itself remains in transition. Affirmative AI endorsements have become more visible in cyber insurance, although their usefulness depends partly on the wording already in the underlying policy. Beazley’s September announcement illustrates that distinction: the insurer said its affirmative AI wording expressly addressed risks already covered under its cyber policies.
Other endorsements extend particular protections. AXA XL’s GenAI endorsement, for example, addresses attacks on third-party data used to train AI models, while its accompanying guidance says attacks on an insured’s own training data were already covered.
Jones views those endorsements as an interim step while insurers work toward a fuller understanding of the exposures they intend to cover. Agentic deployments could become a distinguishing factor in that process.
The liability questions are harder to resolve. There’s no clear answer yet to how responsibility will be allocated when autonomous systems interact and cause harm.
Lockton’s approach with insureds starts with the contractual terms they can control, including liability limits and indemnification obligations. Those provisions also need to be considered alongside each party’s ability to meet its commitments.
However, Jones particularly concerned about businesses providing professional services that rely on AI. She believes deployers could face defense costs and unexpected losses where discussions about faster delivery and revenue growth have outpaced consideration of risk.
That leaves brokers examining both the scope and adequacy of existing insurance. Jones said flexibility in traditional markets, particularly technology errors and omissions, has reduced the incentive for many clients to buy standalone AI coverage. Interest in broader AI liability products has not yet translated into accelerating take-up, in her experience. Available coverage still leaves the question of whether limits are sufficient as deployment expands.
“The soution to these problems should not be insurance first,” Jones said. “Insurance should come after you have assessed and mitigated the risk, otherwise tried to transfer the risk contractually, and then you cede the risk to the carrier through the insurance vehicle, to the extent that is appropriate for that company.”
