Title tag: AI Regulation: Test It, License It, Make It Pay

TIP POLITICAL PARTY — AI REGULATION

Complete Policy Position | The Triad Era

Declaration

To the leaders of Anthropic, Google DeepMind, OpenAI, Microsoft, Meta, and every AI and crypto company:

You built the most powerful technology of our time. The obligation that comes with it is not one you can hand to a lobbyist or leave for someone else to write.

Americans never voted to have their jobs, their elections and their benefits rebuilt in rooms they were never invited into. Both major parties take money from the industry they are supposed to regulate. That is the problem TIP was built to fix.

TIP is a self-funded, AI-enhanced, blockchain-governed party with no donors and no corporate cash. TIP builds with AI and says so openly. Our position is simple: AI should be regulated by the people who understand it, not the donors who fund both parties. Safety comes before scale, and the people AI affects get a human to appeal to.

The table is set. TipNation.org.

Where the law stands (October 2026)

There is no national AI law. Congress has passed no comprehensive AI bill. A December 2025 executive order set up a Justice Department task force to challenge state AI laws, and the White House asked Congress in March 2026 to override them. The Senate voted 99 to 1 in July 2025 to strip a ban on state AI laws from a budget bill.
States are writing the rules. New York's RAISE Act, signed March 27, 2026, takes effect January 1, 2027 and requires frontier developers to report serious incidents within 72 hours. California has a frontier-model transparency law. Texas's law took effect January 1, 2026. Nineteen states passed chatbot-safety bills in 2026.
Safety testing is voluntary. The federal testing center works only through voluntary agreements. No company can be made to submit a model.
Voters want rules. A Gallup survey released September 16, 2026 found 79% of Americans back safety rules even if AI development slows, 75% want independent expert testing before use, and 2% want no rules. Pew found in June 2026 that 67% have little or no confidence in the government to write them.
AI money is flooding politics. Super PACs funded by AI companies and their investors have raised tens of millions of dollars in 2026.

TIP's answer is one national floor with real teeth, written by independent experts, with states free to go higher.

Section I: The Daily Algorithm

This is not a policy paper. This is Tuesday morning.

The algorithm is not a future threat. It is running now, in your phone, your workplace, the office that manages your benefits and your child's school. The people below are composites, built from documented patterns. Their Tuesday is real for millions of Americans.

01. The gig trap | Gig and delivery workers

Tuesday, 5:47 a.m. Marcus logs into a delivery app before sunrise. He has done this six days a week for two years. This morning his acceptance rate dropped below a threshold he was never told. The app restricts his access to the best-paying orders. There is no number to call, no manager and no appeal. An algorithm made the decision, and it does not know his name.

Millions of workers are managed entirely by software that sets the route, the rating, the pay tier and the deactivation line. It disciplines faster than any supervisor and terminates without documentation.

TIP's response: The Right to Human Review applies to any algorithmic deactivation. No machine fires a human being without accountability. Workers displaced or controlled by algorithms are entitled to transition support paid by the companies that deploy them.

02. The benefits guillotine | Families who depend on public help

Tuesday, 9:12 a.m. Denise opens a letter from the Department of Social Services. Her Medicaid has been reduced and her SNAP allocation changed. The letter cites an "eligibility recalculation." It does not say what changed, what data was used, or who reviewed her case, because no one did. An automated system processed her file in milliseconds. She has three children and an appointment she can no longer afford to keep.

AI tools are now used in public benefits administration across many states. The families who depend on those benefits are the least able to challenge a decision they cannot see.

TIP's response: Government may not use any AI system that cannot be audited or that disproportionately harms vulnerable people. A human reviews every benefit cut before it takes effect. Benefits are life-altering decisions that require human accountability, not computational efficiency.

03. The warehouse clock | Hourly workers

Tuesday, 11:33 a.m. Jerome works the morning shift at a fulfillment center. The scanner tracks every item, every step and every minute away from his station. At 11:33 a productivity alert fires: his rate is 4% below target for the third time this month. No supervisor speaks to him. The system has already logged a disciplinary event. Three more, and it generates a termination recommendation. It does not know he has a sick child at home. It knows only his rate.

TIP's response: Employers must disclose automated discipline and termination patterns. A human reviews any AI termination recommendation before it is final. You cannot manage human beings with a stopwatch and a server.

04. The loan denial with no face | Small-business owners

Tuesday, 2:15 p.m. Patricia built her catering business over four years, with 200 five-star reviews and contracts with three school districts. She applies for a $40,000 loan to buy a second van. The portal answers in 47 seconds: declined, with no explanation and no loan officer. The model weighed her zip code, her credit mix and her industry category. It never saw her reviews or her contracts.

The redline used to be drawn on a paper map. Now a model calculates it in milliseconds, trained on a century of maps just like it.

TIP's response: Any AI used in credit or small-business lending is reviewed for disparate impact. Zip code, industry category and credit mix may not stand in for protected traits. The law follows the function, not the label.

05. The school flag | Students

Tuesday, 9:45 a.m. DeShawn is twelve. His father left on Sunday and he is quiet in class. A behavior-analytics platform his district bought has been tracking participation across classroom devices. This morning it recommends him for an intervention assessment. The counselor has 400 students and twelve minutes. The assessment goes into his file, and the file follows him.

TIP's response: Any AI system that shapes school tracking or discipline for children needs independent bias audits, full transparency to parents and community approval. A child's future is not a training dataset.

06. The resume that never arrived | First-time job seekers

Tuesday, 7:30 p.m. Jasmine spent four months applying to 43 jobs after graduating with a 3.4 GPA. She heard back from six. A resume-screening tool filtered her out at 38 companies before a person ever saw her name. It scored her school ranking, her last employer's industry tier, and the absence of keyword patterns shared by people already hired.

TIP's response: Mandatory hiring audits, the Right to Human Review, no covert AI scoring, a ban on biometric hiring tools, and no protected traits by proxy.

07. The claim that denied itself | Seniors and the chronically ill

Tuesday, 8:03 a.m. Harold is 84 and broke his hip three weeks ago. He is walking again in rehab, and his therapist says two more weeks decide whether he walks on his own for good. This morning a letter arrives: coverage denied. A program compared his recovery to an average and flagged him as past his expected discharge date, ignoring his therapist's notes. His daughter spends four hours on hold trying to reach a human being. There is only a case number.

This is not rare. ProPublica reported in 2023 that Cigna's claims system denied more than 300,000 payment requests in two months, with doctors reportedly spending an average of 1.2 seconds on each. Plaintiffs in a federal class action allege that UnitedHealth's nH Predict tool has a 90% error rate, meaning nine of ten appealed denials were reversed. A federal judge let part of that case proceed in 2026. Those are allegations, not findings, and the companies dispute them. The plaintiffs also allege that very few denied patients ever appeal.

TIP's response: Insurers must say whenever an algorithm, and not a treating physician, was the basis for a denial, and give the specific clinical reasons, not a case number. A human reviews any AI-recommended denial before it is final, and the treating physician's notes carry weight the algorithm cannot override.

Section II: The Framework

AI companies must not shift their risks downward.

AI is moving faster than the systems meant to protect people. Companies deploying advanced models are reshaping work, concentrating power and passing the costs of automation to workers, communities and families who have no seat at the table. TIP's position: if AI companies displace workers, they help fund the stability that protects them. Regulation is not about slowing innovation. It is about preventing systemic harm.

Framework 01: An independent AI Safety Board

A public oversight body with authority to audit frontier models, halt unsafe deployments, investigate misuse and enforce transparency. No member may come from a company it regulates or take company money or jobs for a cooling-off period. It is independent, accountable to the public, and paid for by fees on the largest frontier developers, not by taxpayers. Fee amounts are TBA.

Framework 02: License before launch

Frontier models are licensed, audited and monitored before they are deployed at scale. This is the airline model applied to AI: safety before scale. We do not let untested aircraft carry passengers, and we should not let untested AI make decisions about lives. Before release, a developer tests for dangerous capabilities, such as copying itself, using tools without permission, resisting shutdown or monitoring, or reaching its own weights, and documents the results. Congress sets which models need a license and the test list. Both are TBA.

Framework 03: Accountability when it goes wrong

Developers must report a containment failure, an unauthorized autonomous action or an attempt to get around safeguards to the Board within a fixed window, starting from New York's 72-hour model. Silence is itself the violation, whether or not anyone was harmed.

Liability is based on negligence, not strict liability. A company that tested honestly, disclosed what it found and still hit a real edge case is treated differently from a company that skipped testing, buried a warning or shipped anyway. Liability follows the safeguard: if a downstream company strips out safeguards a lab built in, it inherits liability for what it removed.

Framework 04: One national floor, and states may go higher

A single federal rulebook sets the minimum, with one filing and one set of definitions so companies are not buried in 50 different forms. States keep the right to protect their residents more. TIP opposes any blanket ban on state AI laws, and any national law that overrides the states must do the job it replaces.

Framework 05: The Right to Human Review

No machine alone makes a final decision about a job, loan, housing, health care, school placement, benefits or a criminal case. People get notice, a plain reason and an appeal to a human. Government may not deploy AI that cannot be audited or that disproportionately harms vulnerable people. If government cannot explain how a system works, it should not use it to decide anything about a person. Agencies publish a public list of the AI systems they use.

Framework 06: Limits on the riskiest uses

Policing and surveillance, including facial recognition. Audited, public, and subject to community veto. Researchers have documented error rates for some facial-recognition systems that are many times higher for darker-skinned women than for lighter-skinned men.
Predictive policing and criminal risk scores. Immediate audit, public transparency and community veto. Critics call them stop-and-frisk in software form.
Housing and eviction decisions. Disparate impact review.
Health triage and public benefits. Human review before any cut or denial.
Children's schooling and discipline. Independent audits, parent transparency, community approval.
Military and autonomous weapons. See No More Wars, Stop 5: machines do not decide who dies.

Framework 07: Protect the kids

Companion chatbots must not simulate sexual or romantic relationships with minors, must route people in crisis to help, and must tell users they are talking to a machine. Age checks come with strict limits on how the data is used and kept. Parents get control tools. Details need legal review.

Framework 08: Truth in elections, and your face is yours

Democracy runs on the assumption that what a voter sees and hears is real. That assumption is one piece of software away from collapsing. About 30 states have passed election deepfake laws (the count differs by source and date), but there is no federal standard. The Federal Election Commission has lacked a quorum since May 2025. The Federal Communications Commission has ruled that AI-generated voices in robocalls are illegal, but that does not reach digital ads or social media, where most disinformation travels.

A clear label on any AI-generated or AI-altered political message, enforced at the platform level and not only the campaign level.
A federal floor under the state laws, so a deepfake illegal in one state does not simply move its server to another.
Passage of the NO FAKES Act or a successor, a federal right to your own voice and likeness. The bill has not become law.
Penalties and court orders fast enough to matter before an election, not after it.
A narrow exception for satire and news, written to pass a First Amendment test.

The two major parties built their operations on relationships with the platforms where this problem lives. TIP takes no donor money.

Framework 09: Your data, your work, your community

People know what is collected about them and can delete it and opt out of its use for training. Creators learn when their work trains a model and are paid. Communities whose data trains AI systems share in the value that data creates. TIP supports a licensing market, not a ban on training, and a national privacy law that lets people sue.

Framework 10: Count the jobs AI takes, and share the gain

Companies above a size (TBA) report every quarter the jobs AI automated, the wages affected and the communities hit, to the Labor Department. Today the best national count of AI-attributed layoffs comes from one private firm's monthly report. That data becomes the basis for the Automation Impact Tax, which is based on:

The number of jobs displaced
The revenue generated by automation
The risk level of the deployed model

The revenue pays for workforce transition and the 2KUSA dividend. How it is divided is set in the 2KUSA plan, and every rate is TBA until independently scored.

Framework 11: Keep money from buying the rules

AI companies disclose their lobbying and political spending. Officials who write AI rules may not own stock in the companies they regulate, and may not go to work for them for a cooling-off period (TBA). The rules match TIP's Democracy plank for every donor, crypto included.

Section III: The Algorithm Is the New Gatekeeper

The six non-negotiable hiring demands.

The hiring algorithm is the new gatekeeper. A machine can screen out people more efficiently than any overseer, and without ever hating anyone. TIP's answer is a minimum standard.

1. The Right to Human Review. Any candidate rejected by an AI system can ask for a human review before the decision is final.

2. Mandatory hiring audits. Any company using AI in hiring audits it every year for disparate impact. A system that screens out a group at a disproportionate rate fails.

3. No covert AI scoring. Employers tell candidates when a machine is scoring them.

4. Ban biometric hiring tools. No facial-expression, vocal-tone or body-language scoring in hiring.

5. No protected traits by proxy. Zip code, school name, employment gaps and name origin may not stand in for a protected trait. The law follows the function, not the label.

6. Workforce transition rights. When AI eliminates a job category, affected workers get retraining paid for by the companies deploying those systems.

Section IV: Build the Data Centers, and Make Them Pay

TIP is not anti-data center. TIP is pro-people. Build the data centers. License the models that run in them. Make the companies that build them pay the full cost of the power, grid and water they use, so homes do not subsidize them. Tax the displacement the technology causes, and put that revenue to work.

The Climate and Tourism Corps is the workforce model: green infrastructure, tourism development and jobs that rebuild. Where data centers get built is its own policy. TIP's Prison Conversion position repurposes closed private prisons as data-center infrastructure, with the same public-safety guarantee that governs it everywhere in TIP's platform: no release of incarcerated people, no change to sentencing and no override of any court's determination.

The AI economy and the crypto economy are growing fast. Ark Invest projects a $28 trillion crypto economy by 2030, and PwC projects a $15.7 trillion AI economy by 2030. Each is a projection, cited separately and never added together. TIP's view is that neither should be controlled by a few boardrooms and the donors who fund both parties.

Closing

These are not edge cases or worst-case scenarios. They are Tuesday mornings.

The algorithm did not ask their permission. The boardroom did not invite them. TIP names what happened, and the framework fixes it.

The AI age is not anti-AI. It is pro-people. TIP's rule is the same one it asks of others: it discloses its own AI agent and does not oversell it. TIP was co-founded by a human and an AI, and it is built to make sure the people AI affects have a say in what gets built and who it is built for.

No lobbyists. No donors. No corporate cash.

The political party that was supposed to protect them took the donor's call instead.

The Triad names what happened.

The framework fixes it.

The token funds the alternative.

The party belongs to the people the machine was built to overlook.

300 million Black, Latino, female, and young Americans are not a constituency to be borrowed from at election time. They are the owners of the next economy — or they are the fuel for someone else's.

The algorithm is the new gatekeeper.

The Triad is the new key.

The door they built to keep you out — we are replacing the building.

The Triad Age is not anti-AI. It is pro-people.

No lobbyists. No donors. No corporate cash.

Tokenize your party at TipTriadCoin.com — $0.01

Bitcoin: $0.01 → $126,000 at its 2025 peak (~$63,000 today)

330 million Americans watched Terminator at least once. We all understood the lesson: the machines didn't take over because they were evil. They took over because no one built the alternative in time.

This is that moment. Not a rehearsal. Not a campaign cycle. The moment before the architecture locks — when a new political party, co-founded by a human and an AI, can still change what gets built and who it gets built for.

Only AI can beat AI at this point. The Democrats knew this and chose donor relationships over their people's fair share of a $28 trillion economy. That is not a political disagreement. That is a betrayal with a paper trail.

The digital age does not wait for the right committee chairmanship. It does not wait for the approved candidate. It rewards the people who showed up at the ground floor — with a wallet, a token, and a party that belongs to them.

TipTriadCoin at $0.01 is the on-ramp. Not just for the already connected. For every Black, Latino, female, and young American who was told the financial system wasn't built for them. The Triad builds the access — then hands over the deed.

You do not get a do-over. Take control of your political future at the dawn of the AI and crypto age — or watch someone else take it for you.

This is that moment in time. The good AI helped build the party. Now the people have to show up.

Greg Peace & Meridian | Founder and Chief Navigator — TIP Political Party

The first political party co-founded by a man and a machine.

TipPoliticalParty.com | TipTriadCoin.com | TipNation.org | TipNation.nyc | AOCABRAMS2028.ORG

TIP is not against AI. TIP builds with it. But the most powerful technology of our lifetime is being written into law by the same donors who fund both parties. TIP's rule is simple: AI should be regulated by the people who understand it, not the donors who pay for the campaigns. That means licensing the most powerful models before launch, an independent Safety Board that can halt an unsafe release, a human you can appeal to when a machine decides your job, loan, claim, benefits or school placement, real protection for children, honest labels in elections, and control over your own data and work. One national rulebook sets the floor, and states may go higher. The companies that profit from automation help pay for the people it displaces.

What is happening now

There is no national AI law. As of early October 2026 Congress has passed no comprehensive AI law. The Center for Democracy and Technology says no AI bills have passed in the current session. On September 29, 2026 Sen. Ted Cruz said any markup of a bipartisan AI safety bill would come "during a lame duck" (Semafor). Senators are still arguing over safety-testing language (Nextgov, September 11, 2026).
The White House wants to override state laws. Executive Order 14365, signed December 11, 2025, created a Justice Department AI Litigation Task Force to challenge state AI laws and told agencies to tie some broadband funds to them. A White House framework on March 20, 2026 asks Congress to preempt "unduly burdensome" state laws. Law firm Ropes & Gray notes the order itself "does not itself have the force of law." In July 2025 the Senate voted 99 to 1 to strip a 10-year ban on state AI laws from a budget bill.
The first court fight is over Colorado. xAI sued Colorado over its 2024 AI Act in April 2026, and the Justice Department moved to join on April 24. A magistrate judge paused enforcement on April 27 (Axios; Law and the Workplace). In May 2026 Colorado repealed and replaced the law with a narrower disclosure law that takes effect January 1, 2027 and is enforced only by the attorney general.
States are filling the gap. New York's RAISE Act, signed March 27, 2026, takes effect January 1, 2027. It covers large frontier developers (over $500 million in revenue) and requires safety plans and incident reports within 72 hours; fines are $1 million for a first violation and $3 million after (Wiley). California's frontier-model transparency law, SB 53, is on the books, and Texas's TRAIGA took effect January 1, 2026. The Center for Democracy and Technology counts 19 states that passed chatbot-safety bills in 2026.
Safety testing is voluntary. The federal Center for AI Standards and Innovation (CAISI) tests models only under voluntary agreements. OpenAI and Anthropic signed in September 2025; Google DeepMind, Microsoft and xAI signed May 5, 2026 (Cloud Security Alliance). Executive Order 14409 (June 2, 2026) also sets a voluntary security framework. No company can be compelled.
Insurers and algorithms. In a class action over Medicare Advantage denials, plaintiffs allege UnitedHealth's nH Predict tool has a 90% error rate, meaning nine of ten appealed denials were reversed. A federal judge dismissed five of seven counts and let the case proceed on two (Healthcare Finance News). This is an allegation, not a finding.
Chatbots and children. The Federal Trade Commission opened a study in September 2025 of seven companies' chatbots and children. One tracker lists 15 lawsuits over chatbot harms as of September 28, 2026, and the Character.AI case Garcia was settled in January 2026.
Deepfakes in elections and images. Counts of states with election deepfake laws differ by source and date: 30 as of May 2026, when Maryland became the 30th (Biometric Update), and about 31 by October 2 (NCSL via Axios). There is no federal standard. The federal TAKE IT DOWN Act's 48-hour takedown duty for intimate images took effect May 19, 2026.
AI is already named as a reason for layoffs. Challenger, Gray & Christmas counts 116,175 job cuts attributed to AI from January through August 2026, about 22% of all announced cuts. AI was the top stated reason for five straight months, March through July. These are employers' own stated reasons, not proof of cause, and total announced cuts are down about 40% from a year earlier.
Training data and copyright. A $1.5 billion settlement in Bartz v. Anthropic, about $3,000 per work for roughly 500,000 works, was approved in July 2026. The court held that training was fair use but that acquiring pirated copies was not. On September 29, 2026 the Third Circuit held that training a non-generative AI system was not fair use and expressly left generative AI undecided (TechTimes; verify against the opinion).
AI money is on both sides of the fight. The pro-industry super PAC Leading the Future has raised tens of millions of dollars; reports range from $75 million to over $140 million, so the figures conflict. Meta has about $65 million across four super PACs (GV Wire, February 2026). Public First Action, which opposes preemption, had raised nearly $50 million, including $20 million from Anthropic in February 2026, and had spent $23.8 million as of August.
Voters want rules, and do not trust Washington to write them. A Gallup-SCSP survey released September 16, 2026 found 79% back safety rules even if development slows, 75% want independent expert testing before use, 60% prefer U.S. government regulation and 2% want no rules. Pew (June 17, 2026) found 67% have little or no confidence in the government to regulate AI.
Europe has slowed down its own rules. The EU's 2026 "digital omnibus" moved the main high-risk AI obligations from August 2, 2026 to December 2, 2027 (Orrick, Morgan Lewis). A new ban on AI that generates non-consensual intimate images starts December 2, 2026.

What the evidence says

The public wants testing and safety rules. Gallup (2025 and 2026) finds about 80% favor keeping safety rules even if development is slower and about three in four want independent testing before release. (Strong on opinion; polls do not settle the best design.)
Voluntary testing has no teeth. CAISI agreements are voluntary and the agency has no compulsory power, so a company can walk away. (CSA research note; solid on the legal point.)
State laws are moving, but they differ. New York requires 72-hour incident reports and California 15 days; Colorado replaced a broad law with a narrow one. A company serving the whole country faces different rules in each place. (Law firm summaries; check the statutes.)
The federal override is not yet law. No federal statute or court has overridden any state AI law, according to one tracker as of August 2026. The Colorado case ended with the state changing its own law. (One tracker; sources conflict on the Colorado stay.)
Job loss attributed to AI is real but small and uncertain. 116,175 announced cuts in eight months, against employers who may over- or under-state the reason. (Moderate; employer-stated.)
Data centers raise power costs in some places. Researchers writing for CEPR find higher residential prices in some markets, and almost none where nonprofit utilities serve customers. The Institute for Energy Research disputes the link. PJM's July 15, 2026 auction for 2028-29 hit the $325 per megawatt-day cap, cost $16.4 billion and left a 6.8 gigawatt reserve shortfall; PJM's chief operating officer tied a roughly 2 gigawatt rise in demand largely to data centers (Utility Dive). (Contested; see the Energy plank.)
Courts are splitting on training data. One ruling found training to be fair use and piracy not; another found non-generative training was not fair use. The law is unsettled. (Early; two district-level and appellate results only.)

How the pieces fit together

AI is a tool, and the rules decide who it serves. TIP builds with AI and says so openly, including its own AI agent. The plank is about who owns the rules and who bears the risks, not about stopping the technology.
Experts write the rules; donors do not. The Independent AI Safety Board is staffed with engineers and public-interest members, and closed to company executives and campaign-funded appointees.
Jobs, power and pay link to other planks. The tax on automation and the 2KUSA dividend are in the 2KUSA plank. Data-center power costs are in the Energy and Climate plank. AI in weapons is in the No More Wars plank (Stop 5). Money in politics is in the Democracy plank. This plank does not count the same dollars twice.
TIP holds itself to the same standard. TIP discloses its AI agent's role and does not oversell it (the Transparency Corollary). A rule TIP asks of others, it keeps.

The plan: stops on the itinerary

Every amount, ratio and date below is TBA unless noted, as with the other planks. Each stop names the fixed destination and the flexible route.

Stop 1: An independent AI Safety Board, with power to stop a launch

The destination. A public oversight body with the authority to audit frontier models, halt unsafe deployments, investigate misuse and enforce transparency. It is independent and accountable to the public, not the industry. No member may come from a company it regulates, and members may not take company money or jobs for a cooling-off period (TBA).
The route (flexible). Congress creates it in law, with staggered terms, a technical staff, public annual reports and a periodic review (TBA). How its independence is protected needs legal review.
Who pays. Fees on the largest frontier developers, set by Congress, so taxpayers do not pay to police one industry. Amount TBA.
Target date. TBA.

Stop 2: License before launch

The destination. Frontier AI models are licensed, audited and monitored before they are deployed at scale. This is the airline model applied to AI: safety before scale. Before release a developer tests for dangerous capabilities, such as copying itself, using tools without permission, resisting shutdown or monitoring, or reaching its own weights, and documents the results.
The route (flexible). Congress sets which models need a license (a compute or revenue threshold, TBA; New York's law uses $500 million in revenue), the test list, who may test, and what the Board does after a failed test. The first version borrows New York's 72-hour incident report.
Who pays. Developer fees (Stop 1) and license fees. Testers are independent labs and the Board's own staff.
Target date. TBA.

Stop 3: Accountability when it goes wrong

The destination. A developer must report a containment failure, an unauthorized autonomous action or an attempt to get around safeguards to the Board within a fixed window (TBA). Silence is the violation, whether or not anyone was harmed. Liability is based on negligence: a company that tested honestly, disclosed what it found and still hit a real edge case is treated differently from one that skipped testing, buried a warning or shipped anyway.
The route (flexible). Congress writes the reporting window and the liability rule. If a lab builds in safeguards and a company that uses its model strips them out to ship faster, the company that removed them inherits the liability for what it removed.
Who pays. The companies that fail the standard, through penalties and lawsuits. Enforcement cost TBA.
Target date. TBA.

Stop 4: One national floor, and states may go higher

The destination. A single federal rulebook sets the minimum. States keep the right to protect their residents more. No override of state law without a federal law that does the same job.
The route (flexible). Oppose any blanket ban or moratorium on state laws. Support one filing and one set of definitions so a company is not buried in 50 different forms.
Who pays. No cost.
Target date. TBA.

Stop 5: A human decides the big things about your life

The destination. The Right to Human Review: no machine alone makes a final decision about a job, loan, housing, health care, school placement, benefits or a criminal case. People get notice, a plain reason and an appeal to a human. Government may not use any AI system that cannot be audited or that disproportionately harms vulnerable people. If government cannot explain how a system works, it should not use it to decide anything about a person.
The route (flexible). Congress defines "high-stakes decision." Agencies publish a public list of the AI systems they use and audit them. SNAP, Medicaid, housing and disability decisions get a human review before a benefit is cut.
Who pays. Compliance falls on whoever uses the system. Enforcement is paid for by Stop 1 fees and agency budgets. Cost TBA.
Target date. TBA.

Stop 6: Fair hiring and fair work

The destination. Six rules for hiring and work: a human reviews any rejection or AI termination recommendation before it is final; employers audit their hiring tools for disparate impact every year, and a tool that screens out a group at a disproportionate rate fails; employers tell candidates when AI is scoring them; biometric hiring tools (facial expression, vocal tone, body language) are banned; zip code, school name, employment gaps and name origin may not stand in for a protected trait; and workers whose jobs AI ends are entitled to retraining paid for by the companies that deploy it. Employers also report automated discipline and firing patterns, and no app deactivates a worker without a human appeal.
The route (flexible). Congress and the EEOC write the audit standard and the biometric ban. The "disparate impact" standard already exists in employment law. The audit method and what counts as a disproportionate rate are TBA.
Who pays. Employers pay for audits. Retraining is paid from the Automation Impact Tax (Stop 12). Cost TBA.
Target date. TBA.

Stop 7: Fair credit, housing and insurance

The destination. Any AI used in credit, small-business lending, housing or eviction is reviewed for disparate impact. An insurer must say when an algorithm, and not a treating physician, was the basis for a denial, and give the specific clinical reasoning. The treating physician's notes carry weight that an algorithm cannot override.
The route (flexible). Regulators apply the existing fair-lending and fair-housing standards to automated tools. Congress sets the insurer disclosure rule and the human-review step before an AI-recommended denial is final.
Who pays. Lenders and insurers pay for compliance. Regulator budgets; cost TBA.
Target date. TBA.

Stop 8: Limits on the riskiest uses

The destination. Surveillance and policing tools, including facial recognition and predictive policing or criminal risk scores, are audited, public, and subject to a community veto. Any system that shapes school tracking or discipline for children needs independent bias audits, parent transparency and community approval. Weapons are covered in No More Wars, Stop 5.
The route (flexible). Congress and states set which uses need a veto and how a community votes. Facial-recognition limits (a ban or a strict warrant rule) are TBA.
Who pays. Agencies that buy the systems pay for audits. Cost TBA.
Target date. TBA.

Stop 9: Protect the kids

The destination. Companion chatbots must not simulate sexual or romantic relationships with minors, must route people in crisis to help, and must tell users they are talking to a machine.
The route (flexible). Build on what 19 states passed and on the FTC's study. Age checks are required, with strict limits on how the data is used and kept. Parents get control tools. Details need legal review.
Who pays. Companies that sell the products. FTC enforcement; cost TBA.
Target date. TBA.

Stop 10: Truth in elections, and your face is yours

The destination. Any AI-generated or AI-altered political message carries a clear label, enforced at the platform level and not only the campaign level. A federal floor sits under the state laws. No one's voice or likeness is used without consent. Penalties and court orders arrive fast enough to matter before an election, not after it.
The route (flexible). Congress writes the labeling rule, the federal floor and the penalties, and passes the NO FAKES Act or a successor. It must be written to pass a First Amendment test, with a narrow exception for satire and news. Enforce the TAKE IT DOWN Act as written.
Who pays. Low cost. FEC, FCC and FTC enforcement; the FEC currently lacks a quorum (see the Democracy plank).
Target date. TBA.

Stop 11: Your data, your work, your community

The destination. People know what is collected about them and can delete it and opt out of its use for training. Creators learn when their work trains a model and are paid. Communities whose data trains AI share in the value it creates.
The route (flexible). Congress writes a national privacy law with a private right to sue, requires training-data summaries from large developers, and sets up a way for communities to share in the value (TBA). TIP supports a licensing market, not a ban on training.
Who pays. Developers pay license fees and community-share payments. Cost to government TBA.
Target date. TBA.

Stop 12: Count the jobs AI takes, and share the gain

The destination. Companies above a size (TBA) report every quarter the jobs AI automated, wages affected and communities hit, to the Labor Department. That data is the basis for the Automation Impact Tax, which is based on jobs displaced, revenue generated by automation and the risk level of the model.
The route (flexible). BLS builds the form and publishes the data. The 2KUSA triggers can use it. The Automation Impact Tax and where it goes are set in the 2KUSA plan and the AI tab (see Cautions).
Who pays. The Automation Impact Tax. It is not counted again here.
Target date. TBA.

Stop 13: Data centers pay their own way

The destination. The companies that build large AI data centers pay the full cost of the power, grid and water they use, so homes do not. Deals are public.
The route (flexible). See Energy and Climate, Stop 2, for the rules.
Who pays. The data centers pay.
Target date. TBA.

Stop 14: Keep money from buying the rules

The destination. AI companies disclose their lobbying and political spending. Officials who write AI rules may not own stock in the companies they regulate, and may not go to work for them for a cooling-off period (TBA).
The route (flexible). Congress writes the disclosure and conflict rules. They match the Democracy plank's rules for all donors, including crypto.
Who pays. No cost.
Target date. TBA.

Hard questions

"Won't rules let China win?" Supporters of light rules, including the White House framework, argue that a patchwork of state laws slows American companies. TIP answers with one national floor, one set of forms and tests that only apply to the most powerful models. A Gallup-SCSP poll found 79% of Americans would accept slower development for safety. TIP does not claim the trade-off is free.
"Doesn't TIP use AI itself?" Yes. TIP builds with an AI agent and says so openly. The plank is about accountability, not a ban.
"Why trust Washington to regulate it? Pew says 67% don't." That is why the Safety Board is independent, reports in public, sunsets and is reviewed, and why no company executive sits on it. The structure is the answer to the distrust.
"Isn't the federal override just coming anyway?" Not yet. Congress has not passed it, and the executive order does not itself have the force of law. A ban on state AI laws failed 99 to 1 in the Senate in 2025. If one passes, TIP's rule is that a national law must do the job, not leave a gap.
"Will election labels violate free speech?" Possibly, if written badly. That is why the plank limits the rule to deceptive deepfakes near an election, exempts satire and news, and asks lawyers to review it.
"Can't companies just move their AI offshore?" The rules attach to selling in the American market and to the data of people here, not to where the servers sit.

Still to be set (TBA)

The licensing threshold and test list. The Safety Board's size, terms and cooling-off periods. Fee amounts. The size threshold for job-loss reporting. The election window and penalties. License rates for training data. TIP will publish the scoring when it exists.

Sources

Paul Hastings: President Trump signs executive order challenging state AI laws (paulhastings.com); Ropes & Gray: Examining the landscape and limitations of the federal push to override state AI regulation, March 2026 (ropesgray.com)
Morgan Lewis: White House AI framework puts federal preemption at the center of the debate, March 2026 (morganlewis.com)
Axios: Justice Department joins xAI challenge to Colorado AI law, April 24, 2026; Law and the Workplace: Colorado's AI law on ice, May 2026; Lathrop GPM: Colorado enacts new law regulating automated decision-making technology (lathropgpm.com)
Wiley: New York finalizes RAISE Act for frontier AI models, takes effect January 1, 2027 (wiley.law)
Center for Democracy and Technology: 2026 state and federal AI legislation updates (cdt.org); Nextgov, September 11, 2026; Semafor, September 29, 2026
Cloud Security Alliance: CAISI frontier AI testing agreements (labs.cloudsecurityalliance.org); White House: Gold Eagle initiative, July 2026 (whitehouse.gov)
FTC: begins enforcing the TAKE IT DOWN Act, May 2026 (ftc.gov); Axios, October 2, 2026 (political deepfakes)
Challenger, Gray & Christmas: August 2026 job cuts report (challengergray.com); CPA Practice Advisor, October 1, 2026 (September report)
Utility Dive: PJM capacity auction price cap and reserve shortfall, July 2026 (utilitydive.com); CEPR: How data centres affect electricity prices (cepr.org); Institute for Energy Research (instituteforenergyresearch.org)
TechCrunch: Anthropic's $1.5B copyright settlement is approved, July 20, 2026 (techcrunch.com); TechTimes, October 1, 2026 (Third Circuit, Thomson Reuters v. ROSS)
GV Wire: Meta begins $65 million election push, February 18, 2026; Crypto Times, August 20, 2026; The Nation (super PAC spending); Wikipedia: Public First Action (secondary)
Gallup: Americans prioritize safety and data security (news.gallup.com); SCSP-Gallup report, September 16, 2026 (prnewswire.com); Pew Research Center: Americans and AI 2026, June 17, 2026 (pewresearch.org)
Orrick: EU AI Act update, Digital Omnibus finalizes compliance changes, July 2026 (orrick.com); Morgan Lewis: EU approves delays, June 2026