If your claims platform runs AI you have never audited, California now wants to know how you govern it. The state’s insurance regulator has started asking carriers directly about AI, and one question stands out above the rest: who is responsible when the AI belongs to a vendor?
What happened
On November 16, the California Department of Insurance launched a Responsible Use of Artificial Intelligence Survey. It asks insurers how they use AI, what governance sits around it, how much they rely on third-party vendors, and what effect AI has had on their operations. Responses are due December 30. A virtual public workshop on December 15 will cover transparency, fairness, privacy, human oversight, accountability and insurer responsibility for vendor technology. The department says the input will help it decide whether new consumer protections, transparency standards or guidance are needed. (Insurance Journal)
Our take
Most of the survey reads like a standard information request. The workshop agenda is the real signal, because it lists vendor technology as a topic of its own. That matches the direction regulators have taken since the NAIC’s AI model bulletin, which makes insurers accountable for third-party AI systems they use, not just the models they build.
That matters because a large share of the AI in a typical carrier is not built in-house. It comes inside a claims platform, a fraud score, a document-intake tool, a rating engine or an agency system. Many carriers could not produce testing evidence, training-data descriptions or change logs for those models if asked tomorrow. “Our vendor handles that” is not going to satisfy an examiner.
For vendors this is an opportunity as much as a burden. The insurtechs that can hand over clear model documentation and support a carrier’s audit will be easier to buy than the ones that treat their models as a black box.
What to do before December 30
- Build the inventory. List every system that uses AI or machine learning, including features inside vendor products you may not think of as AI.
- Map decisions to models. For each system, note which consumer-facing decisions it touches: pricing, eligibility, claims handling, fraud referrals, marketing.
- Ask vendors for evidence. Request model documentation, bias and performance testing results, data sources and a change-notification process.
- Check your contracts. Make sure you have audit rights, cooperation clauses for regulatory inquiries and notice of material model changes.
- Name an owner. Assign one accountable person or committee for AI governance, so the survey has a clear answer to “who decides?”
California is asking questions now, not writing rules. But surveys like this tend to become templates for other states, and the answers carriers give this year will shape what regulators expect next year.