One Federal AI Rulebook

Enact a national framework for artificial intelligence that displaces the state laws now in force.

Twenty-nine states have passed laws on artificial intelligence and there is no federal statute at all. An attempt to suspend state enforcement for ten years was stripped from the 2025 reconciliation bill by 99 votes to 1; an executive order in December 2025 created a litigation task force to challenge state laws in court instead, and the White House sent Congress a legislative framework in March 2026. That framework asks for a minimally burdensome national standard that preempts unduly burdensome state AI laws, leaves generally applicable state laws protecting children, preventing fraud and protecting consumers in force, and creates no new agency — enforcement would run through the existing sector regulators. The version evaluated here is the statutory one: a single federal standard for transparency, testing and incident reporting, which takes the place of state rules on the same subjects. This evaluation compares five years under that framework against five years of the present patchwork.

Balance

Better for the future · 0.70 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 26 · 70 % Against 11 · 30 %

Size class: small Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 200 million euro per year. This evaluation scores a federal statute on transparency, testing and incident reporting that displaces state rules on those subjects, with child-protection, fraud and consumer laws left to the states as the framework proposes. It does not score the executive order and litigation strategy now in force, which would remove the state rules without putting anything in their place and would therefore score considerably worse. How we score →

Arguments for

Arguments against

6 arguments evaluated · Scoring v1.3 Δ absolute +15

Arguments — For

3 arguments

One rulebook instead of twenty-nine

14of 100

A company deploying the same model in every state currently answers to twenty-nine sets of rules with different definitions of a high-risk use, different disclosure duties and different audit requirements. None of that duplication protects anybody. It is paid for in legal hours.

Value 5 · Enforcement costImpact 3Plausibility 9.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money spent on complying with the same requirement several times over rather than on complying with it once. It is priced at the middle of the scale like any other money and it is a genuine cost rather than a transfer: the legal hours, the duplicate audits and the parallel documentation are consumed. Whether the company or its customers carry it makes no difference to the weight. What the rules themselves achieve is not counted here — this argument is only about paying for the same thing twice, and the value of what would be lost is counted against this measure below. The value is the middle of the scale, the level this site uses for money spent on running a rule.

Twenty-nine states have enacted artificial intelligence legislation and the definitions do not line up: what counts as a consequential decision, what must be disclosed, who must be notified and what an audit has to contain all vary ^[1]^[3]. Roughly 300,000 American businesses deploy artificial intelligence in ways that any of these laws reach — insurers, lenders, employers, health systems and the vendors selling to them. The duplicated compliance cost is put at 2,000 euro each a year, in a range from 700 to 6,000: legal review of each state's requirements, parallel documentation, and the audits that cannot be reused. That gives 600 million euro a year. Taking the two ranges together, short of their joint extremes, the figure runs from about 200 million to 2 billion euro a year. What is not counted is the cost of complying with a single federal standard, which would replace rather than remove the work; only the duplication is counted here. The Impact is the largest on this side and it is the one figure here that follows from arithmetic rather than from a view about what regulation is for.

▸ Show calculation ▾ Hide calculation

	American businesses deploying artificial intelligence in ways state law reaches Setting, range 150,000 to 600,000: no count exists, because the scope differs by state [3]	insurers, lenders, employers, health systems and their vendors	300,000 businesses
×	Duplicated compliance cost each a year Setting, range 700 to 6,000 euro: legal review of each state's requirements, parallel documentation, audits that cannot be reused; below the industry estimates because state templates partly converge	2,000 euro	600 million euro
×	Weight of a euro in company budgets the standard weight this site uses for business money	1.0	600 million euro
÷	Normalised Impact scale of this evaluation	200 million euro a point	3

Score 3 Impact × 5 Value × 9.5 Plausibility ÷ 10 = 14 of 100

That one rulebook instead of twenty-nine saves the duplicated work follows from the law, not from a prediction: once the state rules on these subjects are displaced, nobody has to comply with them. The comparison is the current patchwork, documented state by state ^[3]. What is uncertain is the size, and neither of its two parts has a source: nobody counts the businesses within scope, and nobody publishes what duplicated compliance costs each of them. Comparable estimates for state privacy law vary by an order of magnitude depending on who commissions them, which is why the figure here sits below the industry estimates. Most state laws follow one of two templates, so a company that meets the strictest may already satisfy most of the rest, which would cut the duplication substantially. These doubts sit in the range around the figure, from about 200 million to 2 billion euro a year, and they are not counted a second time here. The remaining doubt is execution: which state rules a federal standard displaces will be argued in court, and some may survive for years. The Plausibility is very high: the saving follows from the displacement itself, and its size is carried in the range.

evidence basis: Mechanism · P ceiling 9.5 identification: Definitional · no rung ceiling

Definitional for occurrence (rule 'occurrence and size kept apart', 02.10.2026): duplicated compliance with displaced state rules falls away once they no longer apply; no behavioural link decides whether it happens. Counterfactual: the current patchwork of twenty-nine state laws, documented state by state [3]. Size: unsourced settings — 150,000 to 600,000 businesses in scope and 700 to 6,000 euro each, plus convergence between state templates — carried in the band 0.2 to 2.0 billion euro, not in P. Enforcement risk (P 9.5 rather than 10): the reach of preemption litigated, with some state rules surviving for years. Direction: not applicable.

Useful systems arrive sooner

8.4of 100

A hospital deciding whether to deploy a diagnostic model, or an insurer a claims model, waits when it cannot tell what the rule will be. Some of what is waiting is worth having, and the delay is not free.

Value 6 · OutputImpact 4Plausibility 3.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is output that happens sooner: diagnoses made earlier, claims settled faster, work done that was not done before. It belongs to the class this site uses for economic systems and prosperity. What is counted is the value of moving a deployment forward rather than the deployment itself, which would happen eventually under either future — the same treatment this site gives to any acceleration. Where the system in question does harm rather than good, that harm is on the other side of this ledger and is counted there. Nothing is priced for the general benefit of artificial intelligence, which is not what this measure decides. The value sits in the middle-upper part of the scale, at the level this site uses for economic output.

American businesses invest something in the order of 200 billion euro a year in artificial intelligence deployment across sectors where state rules bite: health, insurance, lending, employment and public services. Legal uncertainty defers a share of it. A deferral of two percent of that investment by one year is used here, in a range from half a percent to six percent, which is 4 billion euro of deployment moved later. What is lost is not the investment but the return on the delay, put at twenty percent — a high rate, appropriate for a technology whose deployments are expected to pay back quickly. That gives about 800 million euro a year. The figure counts only the acceleration, not the value of the systems themselves, and it counts nothing for deployments that a clear federal rule would prevent rather than enable. The Impact is the second largest on this side and it is the least grounded number in the debate.

▸ Show calculation ▾ Hide calculation

	American investment in AI deployment in sectors state rules reach	health, insurance, lending, employment and public services	200 billion euro
×	Share deferred by a year for legal reasons Setting, range 0.5 to 6 percent: no source; survey evidence is self-reported by firms with an interest in the answer	2 %	4 billion euro
×	Return on a year's delay a high rate, appropriate for deployments expected to pay back quickly; only the acceleration is counted, not the systems themselves	20 %	800 million euro
÷	Normalised Impact scale of this evaluation	200 million euro a point	4

Score 4 Impact × 6 Value × 3.5 Plausibility ÷ 10 = 8.4 of 100

That firms defer irreversible commitments when the rules might change is among the better-established findings in investment economics, and every step after that is assumed here. The comparison is with the current patchwork, with an executive order and a litigation task force adding a second layer of uncertainty on top of it. Nothing measures how much artificial intelligence deployment is actually being deferred for legal reasons rather than for cost, capability or organisational ones, and survey evidence on the question is self-reported by firms with an interest in the answer. The counter-mechanism is serious and unanswered: a federal framework does not end uncertainty if it is contested in court, and the same administration's litigation strategy against state laws is itself a source of the uncertainty this argument wants removed. There is also a real possibility that clear rules accelerate nothing, because the binding constraint on deployment is that the systems do not yet work well enough for the use in question. Nothing suggests the link runs the other way. The Plausibility is low because the deferral this argument prices has never been measured and the counter-mechanism is unaddressed.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain open · P 3–3.5

Counterfactual: the current patchwork, with the executive order and litigation task force adding uncertainty of their own. Design: mechanistic — the uncertainty-to-investment link is carried over from general research; nothing measures deferral in this market, and the survey evidence is self-reported by interested parties. Confounder: capability rather than law being the binding constraint on deployment; unanswered. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds. Band: chain open, because the deferral share carries the whole quantity, has no source, and the capability counter-mechanism is unresolved.

The chain is named but the link carrying the quantity — how much deployment is actually deferred for legal rather than technical reasons — has no source, and the possibility that capability rather than law is the binding constraint is unanswered. Read back: about a third of the time, a clear federal rule accelerates roughly the amount of deployment assumed here.

Open: Deployment timing in states with and without heavy AI statutes, for the same firms and the same use cases, would separate legal deferral from technical readiness and could carry P to 6.

Small developers can afford one rulebook

3.6of 100

The cost of reading twenty-nine statutes is roughly the same whether a company has four employees or forty thousand. That is the shape of a barrier to entry, and it favours exactly the companies whose market position the rules were partly written to check.

Value 6 · CompetitionImpact 1.5Plausibility 4

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is competition: firms that enter a market rather than deciding it is not worth the legal risk, and the pressure they put on the ones already there. This site places it in the class it uses for economic systems and the working order of markets. What is priced is the competitive pressure rather than any particular company's survival, and no weight is given to smallness for its own sake. The compliance cost to the small firms themselves is inside the argument above and is not counted twice. The value sits in the middle-upper part of the scale, at the level this site uses for the working order of markets.

Compliance cost of this kind is close to fixed: understanding what twenty-nine states require takes similar legal effort whatever a company's size, which means it falls on a four-person firm as a large share of its costs and on a large one as a rounding error. The consequence is fewer entrants and more sales to incumbents rather than around them. The value of the competitive pressure lost is put at 300 million euro a year, in a range from 60 million to 1 billion — half the duplication figure above, on the reasoning that the entry effect is real and smaller than the direct cost. This is a price set rather than derived. What runs against it is that a federal framework may itself impose duties heavier than most state laws, in which case the barrier is not removed but relocated. The Impact is half the compliance saving it accompanies, which is the ordinary proportion when a fixed cost is converted into an entry effect.

▸ Show calculation ▾ Hide calculation

	Duplicated compliance cost from the argument above [3]		600 million euro
×	Competitive pressure lost because the cost is fixed rather than proportional Setting, range 10 to 170 percent of the direct cost: the entry effect is real and smaller than the cost itself; nobody has counted the firms that stayed out	50 %	300 million euro
÷	Normalised Impact scale of this evaluation	200 million euro a point	1.5

Score 1.5 Impact × 6 Value × 4 Plausibility ÷ 10 = 3.6 of 100

That fixed compliance costs favour incumbents is well supported across other regulated sectors and has never been measured for this one. The comparison is with the current patchwork. The chain is short and each link is visible: a fixed cost falls hardest on the smallest, the smallest either pay it or stay out, and fewer entrants means less competitive pressure. What is absent is any measurement of the second link — nobody has counted the firms that did not enter, which is the difficulty with every argument of this shape. The counter-mechanism is real and only partly answered: a federal standard is not automatically lighter than the state rules it replaces, and if it is written to the strictest state's level the barrier stays where it is. That depends on drafting rather than on anything empirical. Nothing suggests the link runs the other way. The Plausibility is below the middle: the mechanism is well supported in general, the size is a stated price, and whether the barrier falls at all depends on what the federal standard requires.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: the current patchwork of state rules. Design: mechanistic — fixed compliance costs favouring incumbents is well supported in other regulated sectors, with no measurement for this one and none of the firms that stayed out counted. Confounder: a federal standard written to the strictest state's level, which relocates the barrier rather than removing it; unresolved and a matter of drafting. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds. Band: chain closed but unevidenced — links named, the drafting counter-mechanism stated, only the measurement missing.

Nothing measured argues against the claim; what is missing is any count of the firms that did not enter. The counter-mechanism — that a federal standard could be as heavy as the state rules it replaces — is named and depends on drafting. Read back: about half the time, one rulebook is worth roughly the competitive pressure assumed here.

Open: Entry and financing data for small AI vendors, compared between states with heavy and light AI statutes, would test the entry effect directly and could carry P to 6.

Arguments — Against

3 arguments

The only existing rules would go

9.7of 100

There is no federal artificial intelligence statute. What a federal standard on transparency, testing and incident reporting displaces is the state rules on those subjects — above all the disclosure and appeal rights over automated decisions in hiring, lending and insurance, which are the only ones that exist. The chatbot safety laws for minors are a different subject and are not counted here.

Value 9 · HealthImpact 2.7Plausibility 4

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is what the state rules currently prevent: minors in open-ended companion chats with services that have no safety duty, people refused a job or an insurance policy by a system nobody has to explain, likenesses used without consent. It sits between health and the constitutional core, which is where this site puts a mixture of the two, above money and below life counted alone. The people affected are identifiable rather than statistical — a particular applicant, a particular child — which does not change the weight but does change what the argument is about. What a federal replacement might itself prevent is not counted here, because nothing about it is written yet. The value is high because the stream mixes health with rights that are exercised against a decision, and both sit near the top of the scale.

Twenty-nine states have artificial intelligence statutes of some kind ^[1]. The measure evaluated here displaces the ones on its own subjects — transparency, testing and incident reporting — and the White House framework says in terms that generally applicable state laws protecting children, preventing fraud and protecting consumers stay in force ^[6]; the fourteen chatbot safety laws of 2026 are child-protection rules on a different subject and are therefore not counted here, which is what the first version of this evaluation did by carrying the companion chatbot duties at 0.66 on this scale. What remains are the disclosure and appeal rights over automated decisions in employment, lending and insurance. Some 30 million such decisions a year fall under a state right, in a range from 10 to 60 million; a disclosure or appeal changes the outcome in half a percent of them, in a range from 0.2 to 1 percent — 150,000 people — and a decision corrected is worth about 3,600 euro to the person concerned, in a range from 1,500 to 8,000: a job, a loan, a policy. That is 540 million euro a year, and all of it is counted as lost, because the framework promises a minimally burdensome standard and no text exists; the range runs down to 30 percent lost for a federal text that carries most of the same rights. The Impact is the largest on this side, and it is the one whose size a different draft of the same measure could change completely.

▸ Show calculation ▾ Hide calculation

	Consequential automated decisions a year under a state disclosure or appeal right Setting, range 10 to 60 million: hiring, lending and insurance decisions in the states with such rights; the chatbot safety laws are child-protection rules outside the preempted subjects and are not counted [6] [1][6]		30 million decisions
×	Share where the right changes the outcome Setting, range 0.2 to 1 percent: nothing has evaluated these rights, most took effect in 2026	0.5 %	150,000 people
×	Value of a corrected decision to the person Setting, range 1,500 to 8,000 euro: a job, a loan, a policy	3,600 euro	540 million euro
×	Share not carried by the federal replacement Setting, range 30 to 100 percent: the framework promises a minimally burdensome standard and no text exists [2][6] [2][6]	100 %	540 million euro
÷	Normalised Impact scale of this evaluation	200 million euro a point	2.7

Score 2.7 Impact × 9 Value × 4 Plausibility ÷ 10 = 9.7 of 100

The first half of this argument is a legal fact and the second half is a projection. That preemption removes state rules is what preemption is, and there is no federal statute to replace them; the framework before Congress is a set of recommendations, not enacted text ^[2]. The comparison is the state laws as they stand. Whether removing them costs anybody anything depends on what they achieve, and that needs people to use the rights and companies to change decisions when they do. Here the evidence is thin: most of these rules took effect in 2026 and none has been evaluated. Colorado, which passed the most comprehensive of them, repealed and replaced it in May 2026 with something considerably lighter, which is a genuine counter-argument ^[4]. State laws also differ enormously in bite, and treating them as one quantity hides that. The Plausibility is below the middle: what preemption removes is certain, whether those rules were achieving anything is almost entirely unmeasured.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Rung corrected on 02.10.2026 from definitional to mechanistic (rule 'occurrence and size kept apart'): the removal is definitional — preemption displaces state law by operation of the supremacy clause — but the loss counted, decisions a disclosure or appeal would have corrected, exists only if people use the rights and companies change outcomes, a behavioural chain that nobody has measured. Counterfactual: the state laws as they stand, with no federal statute in force. Design: mechanistic, chain named (right removed → fewer appeals → fewer corrected decisions), unmeasured since most of these laws took effect in 2026. Confounder: state laws differing enormously in bite; Colorado repealing its own comprehensive statute in May 2026 is a counter-indication [4]. Direction: not applicable. Ceiling: mechanistic 6.0; P below it because nothing measured supports the size. The chatbot safety laws are child-protection rules outside the preempted subjects [6] and are not in the quantity.

Nothing measured argues against the claim that preemption removes these rules — that follows from the text itself. What is missing is any evaluation of what they achieve, since almost all took effect in 2026. Colorado repealing its own comprehensive statute is a partial counter-indication and is why the range runs low. Read back: about half the time, the state rules displaced are worth roughly what is assumed here.

Open: The fourteen 2026 chatbot laws and the state automated-decision rules took effect at different dates. Comparing outcomes in early-adopting states against late ones would give the first evidence of what any of them achieve.

Nobody can respond to what's next

1.6of 100

Companion chatbots went from a curiosity to fourteen state laws in eighteen months, and Congress passed nothing in that time. A single federal standard is only an improvement if it can be changed as fast as the technology it governs, and the record suggests it cannot.

Value 6 · Federal latitudeImpact 0.7Plausibility 4

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the capacity to respond to a harm before it is general: fifty legislatures that can each act in a season against one that has not acted in three years. This site places that in the class it uses for economic systems and the working order of institutions. What is priced is the response capacity itself rather than any particular rule it might produce, and nothing here treats state authority as valuable in its own right — the argument is about speed rather than about federalism. The variation between states, which lets a rule be tried before it is imposed everywhere, is part of the same stream and is not counted separately. The value sits in the middle-upper part of the scale, at the level this site uses for the working order of public institutions.

The record here is short and it is unambiguous. Companion chatbots reached mass adoption among adolescents in 2024, California legislated in October 2025, thirteen more states followed within a year, and the federal response as of September 2026 is a committee vote ^[5]. In the same period Congress enacted no artificial intelligence statute at all. What a preemptive federal standard costs is the ability to do that again — but only for whatever falls within its own subjects: the framework leaves generally applicable child-protection, fraud and consumer laws to the states ^[6], so the next chatbot-style response would mostly not be blocked. What is counted is one such episode within the covered subjects in the five years, in a range from none to two, delayed by a year, valued at the scale of the chatbot rules themselves — 660 million euro once, or about 130 million a year across the horizon; the first version carried the full 500 million every year, as if an equivalent response were blocked annually. The Impact is a quarter of what preemption removes directly, which is the right order: losing the ability to make a rule is worth less than losing the rule.

▸ Show calculation ▾ Hide calculation

	Value of the state chatbot rules, as the nearest comparable response [5]	fourteen states in one year against no federal statute	660 million euro
×	Episodes within the preempted subjects over five years, each delayed a year Setting, range 0 to 2 episodes: generally applicable child, fraud and consumer laws stay with the states [6]; delegated agency rulemaking could remove most of the delay [6]	1 episode ÷ 5 years	132 million euro a year
÷	Normalised Impact scale of this evaluation	200 million euro a point	0.65

Score 0.65 Impact × 6 Value × 4 Plausibility ÷ 10 = 1.6 of 100

The premise is documented and the valuation is invented. That states legislated on chatbots within a year while Congress did not is a matter of record ^[5], and the same pattern held for data breach notification, biometric privacy and deepfakes over the preceding decade. The comparison would be a world with a federal standard in place and state authority displaced, which cannot be observed. The chain is short: preemption removes state authority, the next novel harm arrives, nobody can act quickly. The counter-mechanism is genuine and only partly answered — a framework that delegates rulemaking to an agency can move in months rather than years, and the recommendations before Congress contemplate exactly that, though whether an agency would use it is another matter. What has no source at all is the price on a year of delay. Nothing suggests the link runs the other way. The Plausibility is below the middle: the pattern is documented, the valuation is set, and delegated rulemaking could remove most of the concern.

evidence basis: Precedent · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: a world with a federal standard and state authority displaced — not observable. Design: mechanistic — the state-versus-federal speed pattern is documented across chatbots, biometric privacy, breach notification and deepfakes [5], but the price on a year's delay has no source. Confounder: delegated agency rulemaking moving faster than Congress, which the recommendations contemplate; partly answered, since whether an agency would use it is unknown. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds below the praezedenz ceiling of 8.5. Band: chain closed but unevidenced — links named, the delegation counter-mechanism stated, only the valuation unmeasured.

Nothing measured argues against the claim, and the speed difference is documented across four separate technologies. The counter-mechanism — a framework that delegates rulemaking to an agency — is named and unresolved. Read back: about half the time, losing state authority costs roughly a year's delay on the next equivalent response.

Open: Whether the federal framework as enacted delegates rulemaking, and how quickly the agency uses it, settles most of this within two years of passage.

There is nobody to enforce it

0of 100

The state laws are enforced by fifty attorneys general and by private plaintiffs. The federal framework routes enforcement through existing sector regulators and contemplates no private right of action — a thinner apparatus, but not an empty one, and how much thinner is already part of what the replacement is worth in the first argument.

Value 6 · Rule enforcementImpact 0Plausibility 4.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the difference between a rule that is applied and a rule that is written down. This site places enforcement capacity in the class it uses for the working order of institutions, above money and below health. What is priced is the gap between the standard and its application, not the standard itself, whose content is counted in the argument above. Nothing here assumes that state enforcement is vigorous — much of it is not — only that it exists and that fifty offices with the power to act is a different thing from one that has not been created. The value sits in the middle-upper part of the scale, at the level this site uses for the working order of public institutions.

State artificial intelligence laws are enforced through attorneys general and, in several states, through private rights of action that let an affected person sue. The first version of this evaluation read the framework as naming no enforcer; it names no new one, and says so deliberately — Congress should not create any new federal rulemaking body and should work through existing regulatory bodies with subject matter expertise ^[6]. The trade commission's authority over unfair and deceptive practices, the consumer finance and insurance regulators, and the states' own generally applicable consumer and child-protection laws all remain. What is thinner is the absence of a private right of action and of fifty separate enforcers. That difference is real, but it is a statement about what the federal replacement is worth, and the first argument already counts the protection lost net of that replacement. Nothing is added here. The Impact is zero by construction: the enforcement gap is part of the replacement's worth, and that is priced once, in the first argument.

The premise, read again against the primary text, is half wrong: the framework names no new enforcer and routes enforcement to existing regulators ^[6]; that state laws have enforcers is a matter of reading them. The comparison is with the current position, in which fifty attorneys general and, in several states, private plaintiffs may act. What is not established is how much enforcement actually happens: state attorneys general have brought few artificial intelligence cases so far, partly because the laws are new, so the enforcement being lost may be more theoretical than real — which is the strongest counter-argument and is unresolved. The trade commission's existing authority over deceptive practices covers part of the same ground and it opened an inquiry into seven chatbot companies in 2025, which cuts further against this argument. Nothing suggests the link runs the other way. The Plausibility is below the middle: the enforcement gap is real on the face of the documents and how much enforcement is actually being lost has not been established.

Excluded (axis = 0): Two reasons, either sufficient. The framework does name enforcers: it says Congress should create no new AI rulemaking body and route enforcement through existing sector regulators, and it leaves generally applicable state child, fraud and consumer laws with the states [6] — so the premise that nobody is obliged to act was a misreading. And what remains of the gap (no private right of action, sector regulators instead of fifty attorneys general) is a property of the federal replacement, and the replacement's worth is already inside con-1's net figure; pricing it again here counts the same lost protection twice. Set to zero on 14.09.2026.

evidence basis: Mechanism · P ceiling 6 identification: Definitional · no rung ceiling band: Chain closed, unevidenced · P 4–5

Counterfactual: the current position, with fifty attorneys general and, in several states, private rights of action. Design: definitional for the gap — the recommendations route enforcement through existing regulators and contemplate no private right of action [6]. Confounder: state attorneys general having brought few cases so far. Direction: not applicable. Ceiling: mechanistic 6.0 binds for the valuation half. Excluded from the total: the gap is a property of the federal replacement and sits inside con-1's net figure (zero_reason).

The framework routes enforcement through existing regulators and leaves generally applicable state laws in place, so the premise of an empty apparatus does not hold; what remains — no private right of action, fewer enforcers — is priced inside the first argument. Read back: the thinner enforcement is real and is already part of what the replacement is worth.

Open: Whether the enacted framework creates an enforcer and a private right of action is visible in the text, and state enforcement action counts through 2027 would show how much is actually being displaced.

Summary

The case for one federal rulebook is the ordinary case for any of them and it is sound as far as it goes: twenty-nine statutes with different definitions cost money to comply with and that money buys nobody any protection. What the measure costs is narrower than its critics' picture of it. There is no federal artificial intelligence statute and preemption removes the state rules on its own subjects on the day it passes — above all the disclosure and appeal rights over automated decisions, which the framework would replace with a standard it promises will be minimally burdensome and has not written. The chatbot safety laws for minors are child-protection rules on a different subject, which the framework says it would leave with the states, and enforcement would run through existing sector regulators rather than through nobody, though without a private right of action. Whether one rulebook is worth more than what it displaces therefore depends on a federal text that does not yet exist: written to carry the state rights it replaces, it would score clearly better than here; written as lightly as the framework's language suggests, it would score as here or worse — and in either case better than the executive order and litigation strategy currently being used instead, which removes the state rules without putting anything in their place.

Outlook — effect over time

Better for the future · 0.70 previous scale

Sources

White & Case: State AI laws under federal scrutiny: key takeaways from the executive order establishing a federal AI policy framework. whitecase.com
Ropes & Gray: The White House Legislative Recommendations: National Policy Framework for Artificial Intelligence and Federal Preemption of State AI Laws. ropesgray.com
Cloud Security Alliance: State AI Laws Take Hold as Federal Preemption Stalls. labs.cloudsecurityalliance.org
Carpe Datum Law: Colorado's AI Reset: Two Weeks, a White House Callout, and a Pivot Away from the EU Model. carpedatumlaw.com
Transparency Coalition: Watershed year for chatbot safety: 14 new state laws passed so far in 2026. transparencycoalition.ai
The White House: National Policy Framework for Artificial Intelligence — Legislative Recommendations (March 2026). whitehouse.gov
The White House: Executive Order: Eliminating State Law Obstruction of National Artificial Intelligence Policy (11 December 2025). whitehouse.gov

Last reviewed by Claude Opus 5.5 · October 2, 2026 · 3× AI, 1× human

October 2, 2026AI review, approved by a humanClaude Opus 5.5re-scored
Regel ‚Eintritt und Höhe getrennt' (Julian 02.10.) angewendet: pro-1 P 5 → 9,5 mit neuer i_spanne 0,2–2,0; con-1 von definitorisch auf mechanistisch korrigiert (der Schaden hängt an der Nutzung der Rechte), P 4 bleibt; r 0,63 → 0,70, Kategorie besser (ohne Bilanz-Block, pro-2/pro-3 ohne Spanne).
September 14, 2026AI reviewClaude Fable 5.1re-scored
v2 nach Gesamtprüfung 08.09.: Framework-Primärtext gelesen — Kinderschutzgesetze nicht präemptiert, Vollzug über bestehende Behörden; con-1 1,2 → 0,54 (ohne Companion-Anteil), con-2 0,5 → 0,13 (eine Episode statt Dauerstrom), con-3 auf 0 (Fehllesung + Doppelzählung mit con-1). r 0,37 → 0,63.
September 6, 2026AI reviewClaude Opus 5First evaluation
Created for the English side: scores the statutory framework, not the executive order and litigation strategy now in force.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →