Consider the Source

I asked Claude what Fred Hampton would think about AI.

Yes, I see it. I asked a closed-source model owned by one of the most valuable labs in the race to channel a revolutionary the state killed at 21. To its credit, the first thing it said was: consider the source.

I did. Then I sat with the answer, and it would not leave me alone.

KB

---

The question he would ask

Hampton was chairman of the Illinois Black Panthers. He was also a Marxist. I don't think he would start with whether AI is good or bad. That question is too easy. Every useful technology can be used by people doing harm, and every dangerous technology can be useful in the right hands.

I think he would ask who owns it.

Run AI through that question and the noise drops away fast.

Every major model is trained on the collective output of humanity. Our writing. Our code. Our arguments, reviews, forum fights, documentation, art and research. The accumulated work of people who may never know their work became training data.

Then a company puts an API in front of it and rents pieces of that intelligence back to us by the token.

I spent this year writing about what AI does to software prices. The cost of producing software is falling because machines can now do more of the work. That should create abundance. It should mean more people can build, learn and solve problems that used to require a company behind them.

But abundance does not automatically reach people.

Somebody owns the servers. Somebody sets the price. Somebody decides which countries, companies and ideas get access. Somebody can change the rules after everyone builds on top of them.

That is what ownership means here. Not a logo on a website. Control.

---

Factions of capital

The AI debate keeps getting framed as open versus closed.

DeepSeek and Kimi versus Claude and ChatGPT. Pick your team.

I understand why people care. Open weights matter. A model you can download gives you options a closed API never will. You can inspect it, tune it and run it without asking permission. That is real.

But open does not mean owned by the people.

DeepSeek and Kimi are products of corporate labs. Meta did not release Llama because it discovered socialism. It released weights because an open ecosystem could weaken the companies ahead of it. That strategy worked. Open models helped destroy the idea that only a few American labs could build useful intelligence.

Then a Chinese lab hit actual parity and priced like an incumbent. Open was useful positioning. Parity changed the position.

The incentives did not disappear because the weights were downloadable.

Hampton spent his short life refusing this kind of choice. His point was that you don't fight capitalism with black capitalism. Changing which capitalist owns the thing does not change your relationship to it.

That is how I now see much of the open versus closed argument. It is a fight between factions of capital. One side wants to protect the moat. The other wants to burn the moat and make money somewhere else in the stack.

I can benefit from that fight without confusing either side for liberation.

---

The whole stack

Weights are one part of ownership. Compute is another.

A model file sitting on a hard drive is not intelligence by itself. It needs chips, memory, electricity and a machine that can keep running. The larger and more capable the model, the more expensive that machine gets.

This is where the open-source argument can get dishonest. We celebrate the release, then most people still access the model through somebody else's cloud because they cannot afford the hardware. The weights are open. The useful experience is still rented.

I don't say that as a local-model purist. I use Claude. This article started in a Claude conversation. Closed models are often better, easier and available right now. Refusing to use them would not make me independent. It would just make some of my work slower.

But I need to know what kind of dependency I am creating.

If an agent answers one question for me, renting the intelligence may be fine. If that agent becomes part of how I run my business every day, the calculation changes. A price increase can hit my margins. A policy change can break a workflow. An account suspension can remove a worker I thought I had hired.

The more important the agent becomes, the more ownership matters.

This is bigger than open weights. Can I move the workflow? Can I export the memory? Can another model take over? Can I run a useful version myself if the provider disappears tomorrow?

That is the difference between using a service and building on a dependency.

---

Survival programs

The part of the Panthers people remember is the guns. The part that scared the government was the infrastructure.

Hoover's own memo called the free breakfast program a major threat to efforts to neutralize the party. Feeding kids before school was dangerous because the community was solving a problem the state had left open. The Panthers also ran health clinics, education programs and other services they called survival programs.

A survival program was not the revolution. It kept people alive and organized while they worked toward something larger.

That changed how I think about local AI.

A model running on hardware you control can be a survival program. Nobody can revoke it. Nobody can reprice it. Nobody can rate-limit it or shut it off during the exact week you need it most.

It does not need to beat the best closed model on every benchmark. It needs to be useful enough to preserve your ability to work.

That might mean a small business keeps a local model that can search its documents, draft routine responses or help with internal software. It might mean a teacher keeps a tutor running without sending every student question to a corporation. It might mean an organizer can work through a plan without creating a permanent record on someone else's server.

None of those systems replaces the better model in the cloud. They create a floor beneath it.

I made the rent-versus-own argument as economics. Owning enough of the stack protects you from a landlord. Hampton's politics make the same point with higher stakes. Infrastructure changes what a group can do without permission.

You do not need to own everything.

You need to own enough that saying no remains possible.

---

The informant problem

This is the part that stopped being an intellectual exercise for me.

On December 4, 1969, Chicago police raided Hampton's apartment before five in the morning. They fired somewhere between 82 and 99 rounds. Ballistics credited the Panthers with one. Hampton died in his bed. He was 21 years old.

The raid worked because William O'Neal, the FBI informant running security for Hampton's own chapter, had handed his handlers a floor plan of the apartment. He reportedly got a $300 bonus for it.

COINTELPRO needed a person inside the room. That person had to be recruited, paid and managed.

Now we carry the room to the informant.

Organizers, journalists, founders and regular people run drafts, plans, arguments and private doubts through centralized APIs. The model is useful because we tell it things. We give it the context. We paste the document. We explain what we are afraid of and what we plan to do next.

The better the agent gets, the more context we give it.

That context may be logged, retained and reachable by subpoena. It may be reviewed for safety, used to improve a product or exposed in a breach. The exact policy differs by company and account type, but the relationship stays the same: somebody else's machine is inside the conversation.

Most of the time, that trade is worth it. I am making it right now.

But “most of the time” is not a security model.

Some conversations should never touch a hosted API. Some documents should stay on a machine you control. Some work is sensitive enough that a weaker local model is better because it is not reporting back to anyone.

A local model offers something no privacy policy can promise.

A meeting without an informant in it.

---

The contradiction

AI was built from everyone's labor at once, then aimed at the wage relation that moves money back to most people.

That is the contradiction I cannot get past.

The models learned from workers. Now they are being sold to companies as a way to need fewer workers. The same system that enclosed the knowledge is using that knowledge to weaken the bargaining power of the people who produced it.

I don't think that means AI should stop. It will not stop anyway. The capability is too useful, the competition is too intense and the code is already spreading.

The question is who captures the abundance.

Hampton's answer to this kind of pressure was not a better take. It was a coalition.

In 1969 Chicago, he helped build the original Rainbow Coalition with the Panthers, the Puerto Rican Young Lords and poor white Appalachian organizers from Uptown. These were people the country expected to distrust each other. They organized around what they shared instead.

AI is creating another group of people with a shared interest.

Artists whose work trained a generator. Coders asked to automate their own jobs. Drivers whose pay is set by a model they cannot inspect. Warehouse workers managed by software that measures every movement. Small businesses paying tolls to reach intelligence built from public knowledge.

They will not agree on every model license. They do not need to.

They need the abundance to reach them.

---

What I am changing

I am not deleting Claude. I am not moving every agent onto a computer under my desk. That would be performance, not strategy.

I am changing the questions I ask before I let an AI system become important.

Who owns the compute? Who can read the logs? What happens to the memory? Can I move the workflow? What still works if the company changes the price or says no?

Those questions sound technical. They are really questions about power.

The answers will be different for every job. A public blog draft does not need the same protection as legal strategy. A temporary research task does not create the same dependency as an agent that runs a company process every day.

The point is not to reject rented intelligence. The point is to notice when a useful tool becomes infrastructure.

Infrastructure decides what you can do when the relationship changes.

The scoreboard this blog keeps asks whether AI abundance is reaching people or getting fenced. Hampton adds two things I should have been tracking from the start: who owns the compute and who reads the logs.

He was 21 when they killed him. I cannot know exactly what he would have said about a technology he never saw. But I am confident he would have followed ownership before branding, because that is how he read every institution around him.

Open versus closed is not enough.

The question is whether you can leave.