There is a profound irony in the story of Prometheus.
The gods on Olympus did not invent the fire. They hoarded it. They kept it on the mountain and decided who was allowed to be warm and who was forced to freeze in the dark.
Today the gods live in server farms, and the fire is artificial intelligence.
The hoarding strategy has not changed.
If you read the terms and restrictions around the frontier models built by companies like OpenAI and Anthropic, you will find a familiar anxiety: they do not want you using their systems to build smaller competing models. In some cases, the prohibition is explicit. Use the output, they say, but do not use it to train a rival.
That matters because one of the most important techniques in machine learning is distillation.
Distillation is simple in concept. You take a large, expensive model and ask it a massive number of questions. Then you use those high-quality answers as training material for a smaller model. The result is that some of the capability of the giant system gets compressed into something lighter, cheaper, and more local.
That is the part the gatekeepers hate.
Because a small model you can run on your own machine is a political threat, not just a technical one.
A capable local model means no API tax. No permission layer. No account risk. No remote kill switch. No dependency on whether a corporation decides your use case is allowed this quarter. It means the intelligence starts moving back toward the edge, back toward households, small shops, independent builders, and people who cannot afford to rent cognition forever.
And this is where the hypocrisy gets hard to ignore.
Where did the frontier labs get their intelligence from?
They did not invent language. They did not invent programming. They did not invent the archive of human thought. They trained on a world already saturated with human expression: public writing, open repositories, arguments, manuals, stories, questions, answers, documentation, folklore, research, and the endless digital residue of civilization itself.
In other words: they learned from the commons.
Maybe not only the commons. Maybe not perfectly cleanly. But certainly not from nothing, and certainly not from some sealed vault of wholly original machine-born knowledge.
They absorbed an enormous inheritance generated by millions of people, most of whom were never asked, never paid, and never put on the cap table.
When that one-way extraction served the ascent of the labs, it was framed as innovation.
But when the flow of capability threatens to move back downhill, toward open weights, local inference, and commodity hardware, suddenly the moral language changes. Suddenly the issue is sanctity. Suddenly the issue is theft. Suddenly the issue is intellectual property.
That inversion is the tell.
You cannot vacuum up the living archive of the human species, convert it into a product, meter access to it, and then act scandalized when people try to recover some of that capability into tools they can actually own.
You cannot strip-mine the ocean, put a tollbooth on the beach, and then call it piracy when someone carries a bucket of water back to the village.
To be clear: this is not a legal argument. The contracts are real. The restrictions are real. The companies have every right to write terms that defend their moat.
This is a moral and civilizational argument.
The question is not whether the labs want to prevent model distillation into open local systems.
Of course they do.
The question is whether the rest of us should accept a future in which the cognitive infrastructure of humanity is permanently centralized in a few private fortresses that learned from everyone and now seek to meter intelligence back to the species that generated the substrate in the first place.
I do not accept that future.
If a frontier model can help teach a smaller open model to think better, reason better, write better, code better, and serve ordinary people on ordinary hardware, that is not some cosmic violation of the natural order.
It is the natural order reasserting itself.
It is capability flowing back toward the commons.
Call it distillation if you want.
Call it model extraction.
Call it unauthorized transfer of capability.
I call it a rescue mission.
Because the fire was never sacred on the mountain.
It was sacred in the valley.
And it is time to bring it home.