Most of the AI governance conversations in insurance are focused on data, which makes sense.
Data security is a real issue that a multitude of carriers and technologies are about to face as they'll be asked to answer where all data resides, how it's being used and how they came to the decisions they've made in underwriting and claims processing. Those questions need to be asked and answered.
However, there's another AI question coming that people aren't talking about enough yet: what happens when AI usage becomes a resource issue?
AI does not run on magic. It takes infrastructure, computers, processors and electricity. Depending on where and how the data centers are built, their operation can put pressure on energy usage and water supply.
The resource question is only going to continue to become a topic for debate. Insurance is a regulated industry, and we already know what happens when a technology becomes important enough, risky enough or expensive enough that regulators start asking questions. I was onsite with a carrier when HIPAA hit. Privacy went from an abstract topic to a very real operational problem almost overnight.
AI could follow a similar path, but the pressure won't only be about privacy or decisioning. It could be about usage, both in terms of tokens and requests, but also electricity and water. Think about a state like California. They already have water pressure, environmental pressure, and a long history of regulating industries that affect both. What happens if a state starts saying, "You can only use so much AI capacity," or, "You need to report the water and energy footprint behind your AI usage," or, "Certain industries get priority access during constrained periods"?
It's unclear how this will play out, but it's worth carriers and vendors thinking about now. If AI becomes embedded in every workflow, it becomes harder to unwind later. If every claim summary, document review, correspondence draft, payment recommendation, fraud flag, and examiner prompt depends on AI, then AI is no longer a tool sitting on the side but a part of the operating model.
This type of reliance on AI becomes a different kind of risk. The questions start to multiply:
- Could your claims operation still function if AI access were limited, audited or made more expensive?
- Could you decide which AI use cases are essential and which are just convenient?
- Could you explain why one workflow deserves AI support and another does not?
- Could you run the process without it once the usage limit has been reached?
I believe in AI and use it every single day. It removes friction and creates a velocity of productivity that's hard to imagine or match. In the world of insurance claims, AI has the power to help examiners get information faster, reduce repetitive work and ultimately make the experience better for the claimant on the other end.
The strategy will become how to balance the strength of AI with the overuse of AI. Not just for data protection, but for creating a sustainable way to use AI. Carriers will need to create processes to evaluate the use of AI across the organization so that it can be scaled up or down based on their requirements.
Carriers and companies alike won't be able to treat AI like an unlimited utility forever. They should treat it like a powerful resource that needs controls, designing AI usage intentionally. Use it where it materially improves the workflow, but don't just use it because it's available. Keep humans in the loop where judgment matters and track where AI is being deployed to help you understand now which use cases create the most real value, but have a fallback plan.
AI governance is going to be bigger than model governance. Carriers will need to be thoughtful about privacy, security, auditability, cost, infrastructure, environmental impact and operational dependency.
That sounds like quite the laundry list, but insurance has been here before. Regulation always feels abstract until it hits the workflow. The carriers that will be in the best position are the ones asking the next set of questions before they are forced to.
