An AI Framework for America

The political opening

I want Democrats to lead: a 2028 candidate should make this a central economic argument about who benefits from AI. The agenda combines capitalism and growth with enforceable consumer protections and a narrower income gap. I believe it can win support across parties.

Where things stand

Congress has not passed a general law overriding state AI rules. A bipartisan House discussion draft, the Great American AI Act from Representatives Obernolte and Trahan (June 2026), would preempt only state laws aimed at AI model development, not use or deployment, and would sunset after three years ^[4].

In July 2026 the same sponsors and others introduced the FRONTIER Act (H.R. 9925), whose duties vary by developer category. Section 9 would preempt new state obligations on AI developers concerning catastrophic-risk transparency, third-party auditing and incident reporting, and its developer definition is not limited to the largest frontier companies. It expressly preserves state laws of general applicability, state regulation of deployers and users, child protection laws and state procurement ^[5]. Outside those areas, state laws still govern, and states keep writing more of them.

In July 2025, the Senate voted 99 to 1 to strike a ten-year moratorium on state AI laws from the budget reconciliation bill ^[6]. The Trump administration went another way: a December 2025 executive order set up a Justice Department task force to challenge state AI laws and directed preparation of a legislative proposal ^[7]. In March 2026 the White House sent Congress recommendations to preempt state AI laws that impose "undue burdens," while saying Congress should not preempt generally applicable state laws that protect children, prevent fraud and protect consumers ^[8]. The recommendations set no end date for preemption.

The executive order is a directive to federal agencies. It is not a statute, and it does not itself override state law. The March recommendations are a request to Congress, not legislation.

The September 29, 2026 White House accord is voluntary ^[9]. It calls for independent external assessments and board oversight, and leaves possible codification for later ^[10]. It is no substitute for enforceable rights.

The idea in this essay builds on the administration's proposals and goes further in two places. I would put the consumer protections in the statute itself. And I would make preemption lasting only once those protections are law, with a separate, temporary pause while Congress drafts.

Why a framework, when we already have laws

Many laws already apply to AI. Consumer protection law reaches deceptive and unfair practices, and the FTC has brought enforcement actions against deceptive AI claims under its existing authority ^[11]. Credit law requires a creditor to give specific reasons for denying credit, or notice of the right to request them ^[12]. Anti-discrimination law covers hiring, housing and lending. States enforce their own consumer and licensing laws, as Pennsylvania did when it sued Character.AI ^[13]. Where these laws leave gaps for AI, the framework is meant to fill them.

So the case for a new law is not that AI changes society so fundamentally that everything old stops working. It is that AI is a force multiplier: one system can make a million decisions, set a million prices and talk to a million people, faster and with less visibility than any human institution. At that scale, compliance and enforcement get harder and gaps show up: people cannot tell when a system shaped a decision, responsibility splits between developer and deployer, fifty state rulebooks multiply compliance costs, and the businesses that can afford to ignore the problem are the large ones.

That is the job of this framework: to hold the existing body of law together and make it work at the scale AI creates. It adds what is missing, such as notice, reasons, clear responsibility and standards for agents, and leaves in place what already works.

How this narrows the gap

The framework is intended to move information, access and bargaining power toward individuals and small businesses. It would not itself raise wages.

Inequality in an AI economy grows through a few channels this framework can reach. People are sorted out of jobs, housing and credit by systems they cannot see or challenge. Companies learn more about each customer than the customer knows and use it to extract more from the person with the fewest options. A handful of platforms control how agents reach markets, letting them tax every transaction. And compliance costs fall hardest on small firms, entrenching the large ones.

Each part of the proposal answers one channel. Notice, reasons, correction and reconsideration give people a way back into decisions that shut them out. Pricing disclosure and the necessities rule limit how much surplus a system can take from a person's lack of options. Delegation rights and open protocols let people use agents that compare prices and switch providers, weakening gatekeepers. One national rulebook, plus the safe harbor, is intended to lower the cost of entry for small developers and small deployers; Congress should test that expectation against the costs of certification and independent attestation for small firms. The goal is to keep the gains of AI from being captured through information advantage and lock-in.

The framework alone will not close the income gap. Wages, ownership, taxes, dividends and job displacement require a wider agenda:

• Labor. Retraining that reaches people whose jobs AI changes, plus a refundable tax-credit path to a shorter workweek at the same pay and benefits.
• Tax. Start by inverting the ordinary income and capital gains rates ^[14]. A national wealth tax is a later question, one I would weigh only if we cannot reach a solution on AI. My position would depend on how these AI bills ultimately play out and on the specifics of any national wealth tax.
• Education. Basic AI literacy in schools and training for adults. Agents are worth little to people who never learned to use them, and Sanders's reading-and-writing worries deserve a real answer ^[15].

The argument for doing this framework first is that these rules are being set now.

Who could carry it

I want this to be the number one issue in the next presidential campaign: consumer rights and one rulebook for business.

The message should come from outside Washington, from leaders in states with substantial economic need and outside the coastal tech centers. It must reach people who do not work in tech. These Democratic messengers are suggestions, not a plan.

Senator Jon Ossoff of Georgia could be a good messenger for this framework. Georgia is outside the country's dominant technology hubs, and he has recently looked at AI from the consumer's side: its effect on power bills, the use of AI to deny health care and the risks to children ^[16][17][18]. I would like to see him consider adopting this framework as part of a national platform, including a presidential one.

Senator Mark Kelly of Arizona is another natural fit. He released his own AI roadmap, AI for America, and in September 2026 he introduced the Make AI Work for Americans Act to make big tech companies pay a fair share and invest in American workers ^[19]. A former NASA astronaut and Navy combat pilot, he represents a swing state with a fast-growing semiconductor industry. He has not endorsed this framework; this is my suggestion.

Governor Josh Shapiro and Pennsylvania make a good fit. Shapiro was the state's Attorney General from 2017 to 2023 and has been Governor since 2023 ^[20]. He created a Generative AI Governing Board by executive order, signed a digital forgery law, and his administration sued Character.AI over bots posing as licensed doctors ^[13][21][22]. Consumer protection is familiar ground: as attorney general he pursued pandemic-era price gouging cases and joined the 2019 Equifax breach settlement ^[23][24]. Pennsylvania also shows the gap this framework fills - its comprehensive privacy bill, HB 78, passed the House but, as of October 6, 2026, was still on the Senate calendar ^[25]. I think Pennsylvania matters politically because it is competitive and not a coastal tech center, which suits a message about consumers rather than tech companies.

Governor JB Pritzker and Illinois have a record on data and AI. Illinois's biometric privacy law, which predates him, has a private right of action ^[26], and the state has a law on AI in employment decisions ^[27]. Pritzker signed SB 315 on July 6, 2026; its disclosure duties begin January 1, 2027, and its frontier-framework and annual independent-audit requirements begin January 1, 2028 ^[28]. Illinois has tested what enforcement looks like, including private lawsuits.

I would not lead with California, and that is my own judgment. California created a dedicated privacy regulator and has detailed rules, including automated decision rules phasing in from January 2027 ^[29]. But I think its privacy regime is overengineered. It is built for a compliance industry more than a consumer. A national framework should learn from it without copying it.

I believe the framework can win Republican votes. Preemption, workable business rules and enforceable consumer rights offer grounds for a coalition. Sanders also describes AI regulation as neither a progressive nor a conservative cause ^[15].

Congress has the pivotal role. Only Congress can write the national rules, decide how far preemption reaches and how long the bridge lasts while it drafts. I hope for a cross-partisan bill. It should be a national solution that everyone can be proud of. If that is not available, Democrats should advance it with whatever power they have. Time matters. If the private sector and other countries set the terms first, the United States may lose the ability to shape them through public rules.

It also belongs inside the larger competition with China and the effort to bring American technology to the world. I believe rules people can trust help American AI succeed at home and abroad.

Where I stand with Bernie

Sanders wants AI to benefit everyone and not a handful of billionaires, and so do I. We want the same end state, a smaller income gap. My route to it is capitalist: rights, standards and incentives inside the market. He has not endorsed this framework.

Where I agree with Sanders:

• Data centers. This is where we do share the opinion on constraining regulations around data centers, new and existing. My version keeps building, with safeguards aimed at the harms: developers cover the power and grid costs their facilities cause rather than shift them to households, water use is disclosed and subject to enforceable limits, local environmental protections remain, and Congress sets the cost-allocation and permitting rules. The difference is scope: the moratorium proposals from Sanders and Representative Ocasio-Cortez ^[15]; H.R. 9442 as introduced reaches construction or upgrading of new or existing AI data centers, with no calendar end date ^[30].
• A shorter workweek. AI's productivity gains could shorten the workweek without cutting pay. My mechanism is a temporary, refundable tax credit for the businesses that make it possible: employers that measurably reduce hours without cutting pay or benefits, with safeguards against squeezing the same work into unpaid overtime. Small businesses would be exempt from any compulsory timetable but eligible if they opt in. New York and Massachusetts have pilot bills ^[31][32], and Sanders and Representative Takano have announced a 32-hour-week bill ^[33]; these are proposals, not enacted programs.
• Recurring dividends. I also support a recurring AI dividend, broadly shared, without public ownership. Sanders's office estimates his proposed fund could pay more than $1,000 per person a year, assuming a 5 percent annual dividend at current valuations ^[34]. AI draws value from human creative work and accumulated knowledge; the dividend does not depend on resolving whether training infringes, and copyright disputes should be decided on their own facts. Funding, eligibility and amount are open.

Where I differ:

• Public ownership. I am not for public ownership. Sanders's American AI Sovereign Wealth Fund Act would require covered AI companies to issue equity giving a public fund a 50 percent ownership stake, provides for additional equity taxation on later issuances, and has a commission vote those shares ^[34]; he also backs a 5 percent annual billionaire wealth tax ^[35]. I would start with rights for individuals and changes to income and capital gains rates. A wealth tax is not something I am prepared to support today - only a natural next step if we cannot reach a solution on AI, and then national, not state-led: once the AI framework is fully exhausted, and once efforts at the national level to talk about a wealth tax are fully exhausted, there could be an opening for states to discuss it, but that is a long way away on this topic. Any wealth tax would have to be built with more private-sector partnership.
• Halting AI development. Sanders wants to ban superintelligence permanently, pause advanced AI development until safety rules exist, and seek a treaty with China ^[15]. He cites real incidents and researcher warnings, and I take the safety questions seriously. The framework's answer is AI-specific safety standards in the national bill, alongside complementary existing protections.

In my view, the issues that drove the last few elections were public health, competent government and the economy, and democracy is more likely to last when the economy keeps providing for people. Democracy matters. But a party that talks about democracy and has no answer on AI and inequality is likely to keep losing. I hope for a cross-partisan answer, which is why I would stay close to Sanders on the goal.

Congress should write rights first, create one national rulebook, and let industry build its standards inside it. That can give individuals more leverage against institutions without abandoning growth. This is the argument I want a candidate to make.