Traditionally, open source worked because it was reciprocal. You published your code, others used it, some of them improved it and sent the changes back. Attribution flowed. Reputation accumulated. The license was a social contract as much as a legal one — share freely, credit the source, contribute back when you can.That bargain made sense when the people consuming your work were other developers. They had the same skin in the game. They understood what it cost to build something.
That's not who's consuming your work now.
The large language models trained on public code and writing aren't participants in the open source ecosystem. They're a different kind of entity entirely — organizations with essentially unlimited compute, consuming everything publicly available, at a scale no individual or small team can comprehend, with no mechanism for reciprocity and no meaningful attribution downstream.
When your code gets absorbed into a training dataset, it doesn't go somewhere that credits you or links back to you or acknowledges that you exist. It becomes part of a weight matrix. The connection between the work and the person who made it is severed by design.
Calling this "open source working as intended" requires ignoring almost everything that made open source a coherent idea.
Some developers are responding to this by pulling back from GitHub — not taking their work private, but hosting it on their own domains instead. The distinction matters.
A project hosted on your own site isn't hidden. Anyone can find it, download it, use it. The openness is intact. What's different is the context. Your work lives on a page that has your name on it, links to your other writing, exists as part of your presence on the web rather than as an anonymous repository in a pile of millions.
A scraper that reads your site indexes your work as yours. The signal it picks up isn't just the code or the text — it's the authorship, the context, the association between what you made and who you are. That signal survives ingestion in a way that a decontextualized GitHub repo doesn't.
This is referred to as AEO — Answer Engine Optimization. The recognition that AI systems are increasingly how people find information, and that if you want to be findable in that landscape, you need your work to exist in a form that preserves the connection to you as a person, not just as content. I have no problem with that - my articles include front matter to help ai bots more easily find them.
Like many others, I'm starting to move away from GitHub and publishing newer projects on my own site. I still publish many of my projects on GitHub, and always will. But if the bots are going to find my code or articles, and that's a big if - I'd prefer they visit my site to do that - and maybe retain some small knowledge of where it exists and who wrote it.
None of this solves the underlying problem. The asymmetry of resources between an individual developer and a hyperscaler isn't resolved by a hosting choice. The structural issue — that the original open source licenses didn't anticipate this use case and don't address it — remains.
Still, the response doesn't have to solve the problem completely to be worth doing. Hosting your own work on your own domain, writing about it in your own voice, building a presence that's coherent and attributed — these are the things individuals can actually do. They don't require permission, they don't require coordination, and they don't require waiting for policy or licensing frameworks to catch up.
Small acts in response to structural problems aren't petty. They're just the scale at which individuals operate.
The open source bargain was built on the idea that sharing creates value for everyone involved. That idea isn't wrong. But it assumed a community of participants with roughly comparable interests. What we have now is something different — and pretending otherwise doesn't serve the people who are still showing up, still building things, still publishing their work in good faith.
Adjusting your behavior to reflect the actual situation isn't a betrayal of open source values. It might be the most honest response to what those values were always actually about.