A bill in New Jersey has drawn the line this series has been drawing for weeks, in exactly the place that matters. Assembly Bill A5494, introduced September 14, 2026 by Assemblyman Chris Tully, would bar an insurer from using artificial intelligence to make the final decision to deny an automobile, homeowners, or flood claim, with a penalty of up to $5,000 per violation. It is not law. As of late September it sat in committee, with no hearing, amendment, vote, or Senate companion recorded. But AM Best's November 1 roundup flagged it anyway, and rightly so: the bill matters less as pending legislation than as a statement of where the argument is heading.
Read it carefully, because its narrowness is the point. The bill does not ban AI from claims. Triage, assessment, fraud scoring, and approvals can still run on automated tools. It bans one act: the final no. It defines its target narrowly — a data-trained system that simulates human communication and generates non-scripted outputs with limited or no human oversight — and it is silent on partial denials and reduced payments, a gap to notice rather than celebrate.
What it means for carriers
First, the legislature has arrived at the same boundary operators already know. A denial is not a workflow step. It is the moment a carrier breaks a promise in writing, to a named person, with reasons that will be read back in a complaint, a regulator's file, or a courtroom. Everything before that moment can be accelerated. That moment cannot be outsourced to a system that cannot be cross-examined. That is this series' doctrine, now wearing a bill number: agents pre-qualify, a licensed person confirms, the ledger keeps the tape.
Second, the bill exposes how many carriers cannot yet prove where their final no comes from. The question is whether the human who signed it had the evidence, the authority, and the time to disagree. A recommendation that is always approved is a decision, whatever the org chart calls it. A reviewer who sees a score but not the reasoning, the policy language, and the file is a rubber stamp, not a control. A5494's definition will force carriers to map, system by system, which tools fall inside it. Most will find the map messier than their AI inventory suggests.
Third, do not mistake the $5,000 penalty for the price of the risk. Per violation, it is small against the claim itself. The real exposure is what a denied claim reveals on discovery: a final decision from a system nobody in the firm can explain, under a process nobody designed for the denial itself. That is an unfair-claims-practices problem, a discrimination problem, and a trust problem, in that order. The NAIC's examination supplement, covered here last week, already asks whether human review is appropriate rather than merely present. New Jersey is asking the same question as a statute.
From the operator's chair
One: find every final no in your claims operation and name its maker. Not the team, not the platform — the person, or the system, whose output becomes the denial letter. Include the quiet ones: the auto-denial after a deadline, the score threshold that routes straight to a letter, the draft denial approved without opening the file. If you cannot list them, you cannot govern them.
Two: build the denial ledger before anyone asks for it. For every denial, keep the evidence relied on, the reasoning applied, the human who confirmed it, and what that human actually reviewed. If the record cannot show a person exercising judgment, it will not matter that a person technically clicked approve. The ledger is not paperwork for the examiner. It is the proof that the decision was yours.
Three: adopt the standard everywhere, and close the bill's own gap while you are at it. A multi-state book cannot run one denial ethic in New Jersey and another everywhere else without confusing its adjusters, vendors, and examiners. Apply the rule — no AI final denial — across lines and states, and to partial denials and reduced payments too, even though A5494 is silent there. The next bill will not be silent.
Legislators do not write bills like this because they dislike technology. They write them because an industry has not yet shown, in inspectable form, that a person stands behind its hardest decisions. Carriers that can show that — a named human, a reviewable record, a denial that survives being read aloud — have nothing to fear from A5494. Carriers that cannot will keep discovering that the final no was never the machine's to give, and that the cheapest place to learn that lesson is in their own claims operation, before a legislature teaches it to them.
— Krishna