Casar said the U.S. government can’t sit by idly and do nothing. It needs to do something to address the enormous impact of AI on the economy.
AUSTIN, Texas — Artificial intelligence is changing how we live, work and connect. Many people, whether they’re entering the workforce or already in it, worry AI will take their jobs one day.
“Most people involved in artificial intelligence can agree on one thing,” Austin-area U.S. Congressman Greg Casar (D-TX 35) wrote in an op-ed last week. “AI is coming for your job.”
Casar is drawing inspiration from the New Deal–era Works Progress Administration of the 1930s and wants to create an AI tax–funded jobs program to combat potential mass unemployment related to AI.
The New Deal was a series of economic, social, and political reforms launched by U.S. President Franklin D. Roosevelt between 1933 and 1939 to provide relief, recovery and reform following the Great Depression. Casar said that “the kind of mass employment–creating program” like the one FDR created is necessary because AI could create “Great Depression–level unemployment.”
“This AI tax should target the companies that stand to make billions of dollars by laying people off. The tax should be levied on AI providers, not consumers, and it should charge a higher rate for AI used by big corporations rather than individual users,” Casar said. “One goal of this tax should be to shift incentives to save jobs. Our tax system basically gives companies huge tax savings when they automate a job, because we currently tax wages but not AI. If you replace a worker with an AI-powered robot, you save on payroll taxes: That’s functionally a tax break. That’s wrong, and an AI tax would start to change it by leveling the playing field.”
Casar’s plan is to levy a tax on AI tokens, the fundamental units of data that an AI model processes to understand and generate language.
“Tokens refer to the words that are going to be generated by an AI model themselves. So when you prompt ChatGPT, you are asking the model to generate the next-best token. A token is usually a fragment of a word. So the more you are generating, the more you are using ChatGPT or Claude or Gemini, the more tokens you’re using,” Kevin Frazier, the director of the AI Innovation and Law program at the University of Texas School of Law and a senior fellow at the Abundance Institute, said. “You can imagine for a Fortune 500 company, for example, they may need billions of tokens available to their workers to be able to use AI to the fullest extent, whereas a small business owner may need far fewer tokens just to develop a new press release or a new business plan.”
The plan would tax a company’s tokens and put them into separate programs, which Casar said would “slow down job loss” and “fund a jobs program that makes sure we do not leave millions without work.”
Casar said the tax should measure both the number of tokens and the underlying computing power used to train and use AI models, to prevent companies from gaming token counts.
While the response should include job retraining programs, Casar said it should also include direct job creation, because “only funding training at a time of mass unemployment may end up being like offering swimming lessons to passengers of the Titanic.”
“Our goal should be simple: harness the wealth created by AI to create jobs at the same pace AI eliminates them, so that AI does not raise our unemployment rate. If the AI tax rate needs to change to meet that goal, it should be adjusted,” Casar said. “We can look to models like the Universal Service Fund, which requires telephone providers to subsidize telephones for low-income households. That fee is adjusted every quarter to make sure it generates the revenue needed to meet its goal.”
In an interview with CNBC, Casar said the revenue from the AI tax would help create jobs in areas facing worker shortages.
“We have an aging population. We need a lot more elder care in this country. As the father of a very young child, I know we have a huge shortage of child care workers and teachers. Some homes need insulating all around the country, so there’s good work for people to do,” Casar said on CNBC last week. “I think that Americans want careers. They want to help each other out and contribute to this country and so, just like FDR set up the WPA that finished the Hoover Dam, built the LaGuardia airport, I believe that we can come together and have a program that makes sure that Americans stay at work, even in the age of AI.”
Paul Truax, the CEO of Solvis AI, which creates AI products for small businesses, said AI can take some jobs right now, and that will increase over time.
“I think there are scenarios where I could do a lot, if not most, of the tasks of certain jobs. Certain jobs. We’re talking about things like data entry and research. AI is very, very capable of something like that,” Truax said. “There are some jobs that I could do better, and then, if you do the math, if you do the if you look at the economics, it is very much in AI’s favor to do like a certain sector of job.”
In those cases, he does think there should be a tax on that. However, he said AI is not solely responsible for the layoffs and negative economic conditions.
“There’s other factors that are larger, much larger than AI,” Truax said. “For example, the rise in the cost of capital and then also massive over-hiring, specifically in the tech space during the Covid pandemic,” Truax said.
Some critics push back, saying the doom-and-gloom talk of job displacement is overhyped, and that Casar’s proposal could seriously stifle the growth of a rapidly expanding industry.
Frazier welcomed Casar’s creativity and applauded the idea of ensuring work is available to as many Americans as possible, but pushed back on where the funds for a project like this would come from.
“The reality is that we need to be the nation that leads not only in the development of leading AI, but also in the adoption of AI. We have to be the country that knows how to use these tools, that knows how to start new businesses with the latest AI, that makes sure that we are the most AI-savvy people, because that’s where the jobs of the future are going to be,” Frazier said. “Putting a tax on that technology historically hasn’t been a welcomed idea by economists, for example, who fear that such taxes might actually put a cap or otherwise hinder the diffusion and development of this technology.”
Truax said that if the AI industry slowed down, there could be risks from something called “recursion,” in which an AI can improve upon itself.
“If we match that capability with superintelligence, then if an adversary reaches super intelligent recursion five minutes after we do that, AI is already improved upon itself a few hundred thousand times at that point, and this is a hyperbolic example,” Truax said. “That’s the type of risk we are running here where an AI can all of a sudden be hyper intelligent because it reached that recursion before a good player did.”
Truax said he likes where the money from the AI tax would go, to create jobs like caring for the elderly and to support infrastructure. He said there’s enough to be taxed without slowing growth too much.
“If you see the productivity levels that AI can generate, I think that that’s going to lead to enough revenue to withstand a little bit of siphoning off the top, to fund these programs while not slowing down until at least that’s my hope,” Truax said.
Generally speaking, Truax said if we attach a tax on any revenue engine in the AI space, tokenization is likely the best option. Still, not every AI company operates on a tokenized basis, though he said tokens are only going to go up in his view, and that, in the future, everything will be tokenized.
“It’s possible that putting a tax on tokenization could hinder the up-and-coming players, more so than the already big players with deep pockets that can handle that tax,” Truax said. “I will say that that is a concern that I have.”
While he supports the plan, Truax has some questions about exactly how this will be implemented.
“Ultimately it comes down to the details, and it comes down to execution. That is a massive, massive program that he’s proposing, and this is the technology that is changing so fast,” Truax said. “I didn’t see any strict details, any strong details in this plan. I would love to see a little bit more of that.”
Since we don’t know the pace, scale or timing of when these job losses may occur, Frazier said legislators at the state and federal levels are planning for many different futures and should adopt an adaptive framework rather than a heavy-handed or rigid one.
“There have been times during which the nation has seen sudden increases in unemployment, and we’ve historically not responded well to that,” Frazier said. “I think that the proactive posture taken by the congressman should be applauded and is the right approach. What I would caution, though, is reacting without a high degree of adaptability and flexibility.”
Concerns over AI’s impact on jobs are adding to the stress of new graduates entering an already uncertain labor market. At graduations across the country last month, the mere mention of artificial intelligence was met with loud boos.
“AI most certainly is going to change the nature of our labor market, and we have to address that transition period ahead on the jobs that we have today may not be the jobs that we have tomorrow or in five years or in ten years. That’s a real concern for folks who may find themselves on the wrong end of that transition,” Frazier said. “The scale of that, the timing of it is unknown, but the best path forward is to plan for multiple futures such that folks don’t feel like they got caught flat-footed or are otherwise missing out on this AI wave.”
The college class of 2026 faces a tough job market. The unemployment rate for recent college graduates is higher than the national average, and fewer companies are hiring entry-level jobs.
The March unemployment rate among young college graduates ages 22-27 is 5.6%. That’s 33% higher than the unemployment rate among all workers, according to the New York Federal Reserve.
“We’ve seen significant concern around youth unemployment, and this is a pivotal period during which you learn new skills. You receive that training and mentorship from your supervisors, and you go on and start your professional career,” Frazier said. “Making sure that we have a meaningful opportunity for youth to develop those skills is a national priority, or it ought to be.”
In an April poll conducted by the Texas Politics Project, 43% of those surveyed said they are “very concerned” about the impact of artificial intelligence. 32% said they were somewhat concerned, 17% said they were not too concerned, 5% said they were not concerned at all and 2% said they did not know.
The number of people who are very concerned is down slightly from 45% in February but has risen sharply over the last year from 0% back in April of 2025.