New Jersey Bill Would Bar AI From the Final Denial Decision on Home, Auto and Flood Claims

The prohibition in Assembly No. 5494 is one sentence long: “An insurer shall not use artificial intelligence to make the final decision to deny a claim filed under an automobile, homeowners, or flood insurance policy.” Assemblyman Chris Tully, District 38 (Bergen), introduced the bill in New Jersey’s 222nd Legislature on September 14, 2026, as an Act “concerning the use of artificial intelligence in deciding certain insurance claims and supplementing Title 17 of the Revised Statutes.” The text is marked “As introduced.” The Legislature’s bill page records one action on it, dated 9/14/2026: “Introduced, Referred to Assembly Science, Innovation and Technology Committee.”

Section 1 settles which companies the ban covers: “‘Insurer’ means a person engaged in the business of homeowners, flood, or automobile insurance in this State.” Health insurance is not in that definition. Every statute or bill covered on this wire that forbids artificial intelligence from denying a claim has been a health measure, among them the Minnesota law barring automated processing alone from making an adverse determination. Outside health the nearest instrument is Texas Commissioner’s Bulletin B-0003-26, which tells all regulated entities that the Texas Department of Insurance expects a person to review and agree with a consequential AI decision before action is taken. It is an expectation a commissioner published, not a prohibition a legislature enacted.

Section 1 also defines the technology. “Artificial intelligence” means “a technology system that: (1) is trained on data; (2) is designed to simulate human communication through one or more of the following: (a) text; (b) audio; or (c) visual communication; and (3) generates non-scripted outputs with limited or no human oversight.” The three elements are joined by “and,” so on the face of the text a system has to meet all three before the ban touches it. A claims model that reads photographs and a loss description and returns a severity score or a recommended disposition is trained on data and can produce outputs under little human oversight, yet it is not obviously designed to simulate human communication. Whether a system of that kind falls inside this definition is a question the introduced text raises and does not answer.

The statement attached to the bill describes something wider than the defined term. It says artificial intelligence “has been increasingly used in the insurance industry, including in product development, marketing, underwriting and pricing, claim management, automating processes, and promoting efficiency and accuracy.” It then says the bill “makes the claims review process more transparent by banning the use of artificial intelligence to deny claims made under homeowners, flood, and automobile insurance policies and requiring a claims adjuster or investigator to review them.” The review requirement lives in that paragraph. Section 2a bars artificial intelligence from making the final decision to deny and adds no review requirement of its own.

Nothing in A5494 defines “final decision,” and nothing sets a standard for the human step the phrase implies. The ordinary shape of claims automation is a sequence in which a system intakes, scores, ranks or recommends, and a person signs. Where the handoffs in a claim actually sit is the map this language would have to be read against. The text as introduced does not say at which handoff the final decision gets made, or what the signer has to have done for the signature to count.

Enforcement takes two sentences, both in section 2b. An insurer who violates the act “shall be subject to a penalty of up to $5,000 for each violation,” collected “in a summary proceeding in accordance with the ‘Penalty Enforcement Law of 1999,’ P.L.1999, C.274 (C.2A:58-10 et seq.).” Up to, and for each violation: the text sets no floor, no aggregate cap, and no method for fixing the amount. Nothing has to be disclosed to a policyholder or reported to a regulator, no agency is given rulemaking power, and no one is given a right to audit. Three sections in all, the third of which reads “This act shall take effect immediately,” a clause about what would happen on enactment rather than a description of anything in force now.

What sits with the Assembly Science, Innovation and Technology Committee, then, is a one-sentence ban resting on two phrases the bill never settles: its own definition of artificial intelligence, and “final decision to deny.” Read against a working claims pipeline, those two phrases decide most of what there is to know about who the prohibition would touch, and the introduced text leaves both of them to whoever amends, enacts, enforces or construes it.