AI Regulation Tracker / Government and public sector
New York Now Bars State Agencies From Automating Benefits and Rights Decisions Without Human Review
Signed December 21, 2024, the LOADinG Act stops New York agencies from letting software decide benefits, licensing, and other rights questions unless a person stays in the loop and an impact assessment is on file. It has been binding state law since enactment, and the disclosure and assessment clocks have already run.
New York put a hard limit on how its own government can use automation to decide things that matter to people. On December 21, 2024, the Governor signed the LOADinG Act, which became Chapter 674 of the Laws of 2024 and amended the State Technology Law. The name is a mouthful, Legislative Oversight of Automated Decision-making in Government, but the idea is simple. When a state agency uses software to help decide who gets benefits, whose license gets approved, or how someone's rights are treated, a human has to stay meaningfully in charge and the agency has to have done its homework on bias first.
The operative language is blunt. The statute provides that no state agency shall use an automated decision-making system, in the bill's words, "unless such automated decision-making system is subject to continued and operational meaningful human review." That single clause is the whole point. It does not ban agencies from using AI or algorithms. It bans them from handing the decision to the machine.
What decisions does the law actually cover?
The statute reaches automated decision-making systems that are used to make, or to materially assist in making, decisions that affect an individual's receipt of public benefits, their rights, civil liberties, safety, or welfare. That is a wide net. Think eligibility for public assistance, fraud flags that cut off benefits, risk scores that route people through a system, and similar rights-affecting calls. If an agency uses a model to help make that kind of decision, the LOADinG Act applies. It is aimed squarely at the high-stakes government uses of automation, not at back-office tools that never touch a person's rights.
What does an agency have to do before it can deploy?
Two obligations sit at the center. First, continued and meaningful human review. A person with the authority and the information to actually override the system has to remain in the loop for the life of the deployment, not just at rollout. Second, an impact assessment. Before using a covered system, and then on a recurring basis, the agency has to complete an assessment that documents the system's purpose, its development, the data and logic behind it, and, critically, testing for bias and discrimination against protected characteristics. If that assessment shows the system produces discriminatory or biased outcomes, the agency has to stop using it. The assessments are not one-and-done. They have to be redone at least every two years, and existing systems in use when the law passed had to be disclosed to the legislature within one year of enactment.
Read the status precisely: this is a binding statute
I want to be exact here because a lot of AI policy news is aspirational and this is not. The LOADinG Act is signed law. It is enforceable, it amends the State Technology Law, and the timelines in it have already started running. This is not a task force, a set of principles, or a bill still moving through committee. If you run a New York State agency, or you sell automated decision tools to one, the requirements are live obligations, not a preview of what might come.
What this means for executives and their vendors
For executives inside New York State government, the move is to inventory. Know which of your systems make or assist rights-affecting decisions, confirm each one has a current impact assessment that tests for bias, and be able to name the human who reviews and can override the output. A tool without a completed assessment is a tool you are not supposed to be running for covered decisions. For vendors, the calculus has changed. Selling an automated decision system to a New York agency now means your buyer needs assessment-ready documentation, explainability of the underlying logic, and a governance story about human oversight. If your product is a black box that makes the call on its own, it does not fit what the statute allows, and your agency customer cannot lawfully deploy it for covered uses. The commercial reality is that documentation and human-in-the-loop design are no longer nice-to-haves in this market. They are the price of entry.
Questions professionals are asking
Is the LOADinG Act actually in effect?
Yes. The Governor signed it on December 21, 2024 as Chapter 674 of the Laws of 2024, amending the State Technology Law. It is binding, enforceable law. The disclosure deadline for existing systems ran one year from enactment, and impact assessments must be refreshed at least every two years.
Does it ban state agencies from using AI?
No. It bans using an automated decision-making system to decide rights-affecting matters without continued meaningful human review and a completed impact assessment. An agency can use AI to inform a decision, but a person with real authority has to stay in the loop and be able to override the output.
Which decisions are covered?
Decisions that affect an individual's receipt of public benefits, rights, civil liberties, safety, or welfare. That includes eligibility determinations, benefit terminations, risk scoring, and similar high-stakes government calls. Back-office tools that do not touch a person's rights are outside the core of the statute.
What does the required impact assessment involve?
The agency documents the system's purpose, development, data, and underlying logic, and tests it for bias and discrimination against protected characteristics. If the assessment shows discriminatory or biased outcomes, the agency must stop using the system. Assessments must be redone at least every two years.
What should a vendor selling to New York agencies do?
Expect your agency buyer to need assessment-ready documentation, explainability of the model's logic, and a human-oversight design. A black-box tool that makes covered decisions on its own cannot lawfully be deployed for those uses, so build the governance and documentation in before you pitch.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how the LOADinG Act applies to your agency or product with qualified New York counsel.