← Back to News

Where is the Prison Labor Market Headed?

6 min read

I am not an economist. I am a founder who built a company a year into my own recovery, and I spend most weeks talking to the people who run reentry programs. So this is not a policy paper. It is a set of questions I cannot stop asking, and I think you should be asking them too.

Because two facts are about to collide.

The first: artificial intelligence runs on human labor. Every model you have ever used was shaped by people reading text, labeling images, ranking answers, and correcting mistakes, hour after hour. The industry needs millions of hours of human judgment, and it has always gone looking for the cheapest fluent people it can find.

The second: American prisons hold a captive workforce. About two thirds of the 1.2 million people in state and federal prisons work. They produce close to $11 billion in goods and services a year. They earn, on average, between 13 and 52 cents an hour. In seven states, including mine, most of that work is not paid at all. Up to 80 percent of the wages that do exist are taken back for room and board, court costs, and fees. Seventy percent of incarcerated workers say they cannot afford basic necessities on what they earn. Seventy percent say they received no formal training for the job. (Source: Captive Labor, ACLU and the University of Chicago Global Human Rights Clinic, 2022.)

Read those two paragraphs together and tell me you do not see what is coming.

Where it has been

The Thirteenth Amendment abolished slavery and involuntary servitude, except as punishment for a crime. Many states treated that exception as an instruction. Convict leasing followed, and the South, my part of the country, ran on it for decades. The leases ended. The logic did not. Today the prison itself is the main employer, and the work is mostly the maintenance of the prison: the kitchen, the laundry, the grounds. The value is real. The paycheck is not.

So here is the first question. If the work has value, where does the value go?

Where it is headed, version one

This is not hypothetical. It has already happened.

In Finland, a market data company called Metroc needed people who read Finnish to label construction documents so its AI could learn the language. Finnish speakers are expensive. So the company took the idea to the prison service. Incarcerated people were given laptops and asked simple questions about snippets of text, in three hour shifts, for about 1.54 euros an hour. The same rate as any other prison job there. The program was voluntary, framed as rehabilitation and digital skills, and the prison service itself said the purpose was not to create data for companies and that it could not expand very far. (Reporting: Wired, Euronews, 2023 to 2024.)

Finland is the humane version. Voluntary. Paid the same as everything else. Explicitly capped.

Now imagine the same idea in a system where a majority of incarcerated workers say the work is mandatory, where the wage is zero in seven states, where the tasks are monitored, short, and easy to move somewhere cheaper. Imagine the AI industry discovering that the most captive fluent workforce on earth is sitting inside American prisons, already legally exempt from minimum wage, overtime, and the right to organize.

Would you call that training? Would you call it a job?

If the pay is zero, what exactly did the person learn about the modern workplace, other than that it does not pay them?

Where it is headed, version two

There is another fork, and I have seen it.

Organizations are already running coding labs inside correctional facilities, teaching real skills on real equipment and placing graduates in real jobs on the outside. Persevere does this across multiple states and reports a placement rate above ninety percent. The Last Mile has been doing it since San Quentin. Nobody in those rooms is labeling data for cents. They are learning the thing the outside pays for.

Congress created a program decades ago that allows private employers to hire people inside at the prevailing wage for the work, with deductions that still send money to victims, families, and the state. It is small. Almost nobody has heard of it. Why?

The Finnish system runs on something they call the normality principle: conditions inside should look as much like life outside as possible, because that is what people are being prepared for. Apply that principle to AI and the answer is obvious. If working alongside AI systems is what the modern workplace looks like, then people inside should be working alongside AI systems, on real tools, for pay that resembles the outside, with a skill that leaves with them.

That is not a labor pitch. That is a reentry strategy.

My actual position

Here is the thesis, plainly. The infrastructure decides which fork we take.

Nothing runs inside a facility without secure, contained deployment: managed devices, allowlisted networks, audit logs, safety systems that a warden and an IT director will both sign. Whoever builds that infrastructure first decides what runs on it. The same locked-down laptop can deliver a data labeling queue at zero dollars an hour, or it can deliver a curriculum, a certification, a real tool, and a paid role that teaches someone how to work with the machines that are reshaping every job they will ever apply for.

Same hardware. Opposite futures.

I want the second one built faster than the first one, and I am not neutral about it. That is why constrAInt is building facility mode with the organizations that run programs inside, and why I would rather say the next sentence out loud than let anyone wonder: we are not building a marketplace for prison labor. We build practice. The interview. The resume. The hardest conversations someone will ever have about their own past, rehearsed in private, where failing is free. If real work comes inside, and it should, our job is to make sure the person can walk out and turn that work into a career.

What fair looks like to me is simple. Voluntary. Paid like the outside for the same work. A skill that transfers. Value that follows the person out the door as savings, not room and board.

The questions I cannot answer for you

If a task is worth twenty dollars an hour outside the fence, what is it worth inside? Who decided?

If the AI industry needs a million hours of human judgment this year, who should get paid for it?

When someone walks out, what do they leave with: a skill, a certificate, a savings balance, or a bill?

Would you take the job if you were allowed to quit?

If your state pays nothing for prison work today, what does it plan to pay for AI work tomorrow?

Who benefits when a program like Finland's cannot scale? Who benefits when it can?

And the one underneath all of them. When AI comes inside, and it is coming, who does it come inside FOR?

I wrote last month that a job is not enough, and that the goal is not employment, it is a career that pays the rent. The prison labor market is about to get a new employer. Whether that employer teaches or extracts is not decided yet.

It gets decided by what we build.

For the Hardest Conversations in Employment.