The Preemption
When Congress preempted every state AI law in the name of innovation, Maya Chen thought it was just politics. Then the automated termination notices started arriving — and nobody had the authority to stop them.
The office had been empty for eleven months, but Maya Chen still dreamed about it.
In the dream, she was always sitting at her desk on the fourth floor of the Ralph L. Carr Colorado Judicial Center, reviewing a complaint. The details shifted each time — sometimes it was a hiring algorithm rejecting every applicant over forty, sometimes a predictive policing system flagging the same block in Aurora over and over — but the feeling was always the same. She had authority. She could pick up the phone, open an investigation, issue a cease-and-desist. The machinery of state government, slow and imperfect as it was, could grind toward something resembling accountability.
Then she would wake up, and remember that the machinery had been disassembled.
The American Innovation and Technology Leadership Act — AITLA, which the press had briefly and then permanently shortened to "the Preemption" — passed in June 2027 with bipartisan support. Sixty-three senators. Two hundred and eighty-one representatives. The President signed it on the South Lawn with the CEOs of nine major technology companies standing behind her, smiling in the summer heat.
Section 4(a) was the blade. "No State or political subdivision of a State may establish, mandate, or enforce any law, regulation, ordinance, or requirement relating to the development, deployment, or operation of artificial intelligence systems." Eighteen months of regulatory architecture in Colorado — the Algorithmic Accountability Act, the Automated Decision Systems Registry, the AI Impact Assessment Framework that Maya had spent two years helping to build — erased in forty-one words.
Maya's position was eliminated in August. The entire Office of AI Compliance was dissolved. Her supervisor sent a brief email. "Our enforcement mandate has been federally preempted. I'm sorry. You did good work." He was gone by the following week, hired by one of the companies he'd once investigated.
The federal replacement was called the National AI Governance Board. It had a website, a logo, and an eleven-member commission that had met twice in fourteen months. Its complaint portal accepted submissions. Maya had checked — there were no public records of any complaint being resolved. The Board's sole enforcement action to date was a nonbinding letter of guidance sent to a company that had already discontinued the product in question.
This was the world in December 2028, the world Maya was trying to explain to Ruth Gallegos, who was sitting at Maya's kitchen table and quietly falling apart.
Ruth lived three doors down. She was sixty-seven, a retired school librarian with a paid-off house and a vegetable garden that produced more zucchini than the entire block could consume. Maya had always known her as steady. Unshakable. The kind of woman who brought soup when you were sick and remembered the names of your cousins.
Now Ruth's hands were shaking so badly she couldn't hold her coffee mug.
"It started with the insurance," Ruth said. "Three weeks ago. I got a letter — not a real letter, one of those automated things — saying my homeowner's policy was being 'adjusted' based on a comprehensive risk reassessment. My premium tripled. Tripled, Maya. I called them. I was on hold for two hours, and when I finally got through, the woman said the decision was made by their risk modeling system and couldn't be manually overridden. She said I could file a dispute, and the dispute would be reviewed within ninety business days."
Maya nodded. She had heard variations of this story before, from former colleagues, from posts in the regulatory listservs she still followed. She knew what was coming next.
"Then the credit card," Ruth continued. "Visa. The one I've had for twenty-two years. Canceled. Not reduced — canceled. The letter said my 'financial behavior profile' had been updated and I no longer met the criteria for the product. I have never missed a payment. Not once in twenty-two years."
"Did you call them?"
"I called. They said the same thing. Automated decision. Dispute process. Ninety days." Ruth paused. "Then last week, my bank froze my checking account."
Maya felt the familiar nausea. Not surprise — she had lost the capacity for surprise about these things months ago. Just the dull, grinding sickness of recognition.
"They said there was a 'fraud risk indicator' on my profile. They couldn't tell me what the indicator was. They couldn't tell me where it came from. They said they were legally required to act on it, and that I could file a dispute, and the dispute would be..." Ruth trailed off.
"Reviewed within ninety days," Maya finished.
"I can't buy groceries." Ruth's voice cracked. "I can't pay my electric bill. I went to the food bank on Tuesday. I've never been to a food bank in my life."
Maya reached across the table and took Ruth's hand. "I'm going to help you. We'll figure this out."
She said it because it was what Ruth needed to hear. But she knew, with a certainty that sat like a stone in her stomach, that there was almost nothing she could do.
The next two weeks were an education in the architecture of helplessness.
Maya started with the Colorado Attorney General's office, where she still had contacts. Her former colleague, Priya Okafor, answered on the second ring and listened with the weary patience of someone who had heard the same story dozens of times.
"Maya, I believe you. I believe everything you're telling me. But we have no jurisdiction. AITLA preempted our enforcement authority over AI systems completely. We can't investigate, we can't subpoena, we can't issue orders. We tried — remember the Pinnacle Health case in September? The federal court shut us down in three days. We're not even allowed to send a formal inquiry."
"What about consumer protection? General fraud statutes?"
"The companies aren't committing fraud. They're making automated decisions based on their models, which is explicitly protected activity under AITLA Section 7. We looked at this from every angle. There's nothing."
Maya tried the National AI Governance Board next. The complaint portal was a single web form: name, email, description of issue, affected AI system (if known), supporting documents (max 5MB). She spent three hours writing a detailed complaint on Ruth's behalf, documenting the cascading failures and the apparent coordination between systems sharing data through what she suspected was a common risk-scoring platform.
She received an automated acknowledgment. "Your complaint has been received and assigned reference number NAIGB-2028-447291. Complaints are reviewed in the order received. Current estimated review time: 14-18 months."
Fourteen to eighteen months.
She tried calling. The Board's phone number connected to a menu system with seven options, none of which led to a human. Option 4 asked for the reference number and played a recording: "Your complaint is in the queue. No additional information is available at this time." Option 7, "Speak to a representative," played hold music for forty-five minutes and then disconnected.
Maya called her congressman's office. A legislative aide named Tyler took notes and expressed concern. "We're hearing a lot about this kind of thing," Tyler said. "The congressman is very aware of the issue. He's co-sponsoring the AI Accountability Restoration Act, which would give states limited enforcement authority in cases of demonstrable harm."
"When will that be voted on?"
"It's in committee. Realistically? Maybe next session. There's a lot of pushback from the tech caucus."
"My neighbor can't buy food."
"I understand, ma'am. I can flag this as an urgent constituent case. Let me give you the number for the FTC complaint line."
The FTC complaint line was another web form. Another automated acknowledgment. Another reference number vanishing into the administrative void.
It was on the third week that Maya found the loop.
She had been researching the companies involved — the insurance provider, the credit card issuer, the bank — and discovered that all three used risk assessment data from the same third-party platform: Clarion Analytics. Clarion's website described it as "the nation's leading AI-powered consumer risk intelligence provider." It aggregated data from public records, financial transactions, social media, and what it called "behavioral inference modeling" to generate a unified risk score for every adult in America.
Clarion also, Maya discovered, operated the dispute resolution system that all three companies directed their customers to. When Ruth filed a dispute with her insurance company, her bank, and her credit card provider, all three disputes were routed to Clarion's Automated Resolution Engine, which reviewed the disputes against the same risk model that had generated the adverse decisions in the first place.
The system was reviewing its own work. And finding, with perfect consistency, that its work was correct.
Maya felt something cold move through her chest. She pulled up Clarion's corporate filings and found that their Automated Resolution Engine had a 99.7% rate of upholding initial risk determinations. They published this statistic proudly on their investor relations page. "Industry-leading consistency in risk assessment outcomes."
She dug further. Clarion had a human appeals board for the 0.3% of cases that made it past the automated review. The board consisted of three employees. They were responsible for reviewing appeals from an estimated forty million adverse decisions per year. Maya did the math. Even if every member of the three-person board worked every business day, they could review roughly one appeal per minute. At that rate, they would need over seven hundred and sixty years to clear a single year's caseload.
The appeals board was not a real thing. It was a line item on a compliance document, a fiction designed to satisfy the minimal federal requirements. The loop was closed. The AI made the decision. The AI reviewed the decision. The human appeal was a statistical impossibility. And no state agency in America had the authority to pry it open.
Maya sat at her desk, staring at the numbers. Then she called Ruth.
"So there's no one," Ruth said. It wasn't a question.
"The federal complaint is filed. The congressman's office has it flagged. I've documented everything — the data sharing, the circular dispute process, all of it. I'm going to send it to every journalist I can find."
"But there's no one who can make them stop."
Maya opened her mouth. Closed it. "Not right now. Not with the current law."
Ruth was quiet for a long time. Through the phone, Maya could hear the sound of the television — Ruth had started leaving it on all day, she'd mentioned, because the silence in the house had become unbearable.
"My daughter in Phoenix wired me money," Ruth finally said. "For groceries. She had to wire it because my bank account is still frozen. I had to pick it up at a Western Union counter. I felt like —" She stopped. "I spent thirty-one years teaching children how to research, how to find answers, how to navigate systems. I was good at it. And now I can't navigate my own life."
"This isn't your failure, Ruth. The systems are designed to be unnavigable."
"That doesn't help me buy groceries."
No. It didn't.
Maya published her findings on her blog — the one she'd started after losing her job. She wrote about Ruth's case in detail, with permission, using anonymized identifiers. She laid out the circular dispute architecture, the three-person appeals board, the 99.7% uphold rate. She tagged every journalist, every policy researcher, every former regulator she knew.
The post got nineteen thousand views in two days. Several journalists reached out. A policy institute cited it in a report. For forty-eight hours, Maya felt something she hadn't felt in months — the dim, fragile sense that bearing witness might matter. That documentation could be a form of power.
On the third day, she received an email from Clarion Analytics' legal department. It was polite and brief. It informed her that certain statements in her blog post constituted "materially misleading characterizations of proprietary risk assessment methodologies" and that continued publication could expose her to liability under the AITLA Section 12 protections against "interference with authorized AI operations." It requested that she remove the post within seventy-two hours.
Maya did not remove the post. She shared the legal threat on her blog, too. That got another twelve thousand views.
She felt, briefly, brave.
The letter arrived on a Monday.
Not an email — a physical letter, printed on heavy stock, delivered by certified mail. Maya signed for it at the door and carried it to the kitchen table, the same table where Ruth had sat six weeks earlier with her shaking hands and her cold coffee.
The return address was her mortgage company.
She opened it and read the first paragraph.
"Dear Maya Chen: Following a comprehensive reassessment of your financial risk profile by our automated underwriting system, we regret to inform you that your mortgage terms are being adjusted effective immediately. Your monthly payment will increase from $1,847.00 to $4,291.00, reflecting updated risk factors associated with your current profile. If you are unable to meet the adjusted terms, foreclosure proceedings will initiate automatically after ninety (90) days of non-payment."
Maya set the letter down. She read it again. The words didn't change.
The second paragraph directed her to the dispute resolution portal. She already knew which company operated it. She already knew the 99.7% uphold rate. She already knew about the three-person appeals board and its seven-hundred-and-sixty-year backlog.
She turned the letter over. On the back, in small gray print, was the legally required disclosure: "This decision was generated by an automated system. For questions about federal AI governance standards, visit www.naigb.gov or call 1-800-555-0147."
Maya stared at the phone number. She had memorized it weeks ago. She knew exactly what would happen if she called — the seven options, the hold music, the disconnection. The loop.
Outside, a lawnmower droned in a neighbor's yard. A dog barked. Somewhere down the street, Ruth Gallegos was probably sitting in her living room with the television on, waiting for a resolution that existed nowhere in the architecture of the world they now inhabited.
Maya picked up the letter again. Her hands, she noticed, were perfectly steady. That was the strange part. She had expected fear, or anger, or the frantic energy of someone who still believed there was a door to knock on, a person to call, a lever to pull that would make the machine acknowledge she was a human being and not a risk score.
Instead she felt something quieter and more terrible. Recognition. She had mapped this system. She had documented its impossibility. She had explained, in clear and careful prose, exactly how the loop worked and why no one could break it.
She just hadn't understood, until this moment, that she was inside it.
The lawnmower droned on. Afternoon light came through the kitchen window at its usual angle. Everything was ordinary. Everything was working exactly as designed.
Maya set the letter on the table, placed her hands flat on either side of it, and sat very still in the quiet of her house, waiting for nothing.