Thoughts on my code being part of AI training repo

Hugging Face recently released The Stack v3 as a massive dataset for training LLMs to write code. From Hugging Face:

The Stack v3 is a 15.9 TB dataset of source code across 713 programming languages from 173M repositories, crawled from GitHub in 2025.

Hugging Face provides a search for finding codebases and a way to remove them. Three of my ColdFusion codebases are included: bear, codey, and improveApp. It turns out I released only one of them under the MIT License, but it certainly was my intention for all three.

I made the repos public as a contribution to the wonderful ColdFusion community, hoping they’d help other engineers understand ideas, work faster, and write better code.

Is it okay for AI companies to use this?

I’m fine with it, and here’s why.

The premise of the MIT License is that anyone can take the code and use it, including in commercial software. Does that extend to companies using it to train LLMs? That’s a different situation from authors who expected to be paid for their work, which is the premise behind the lawsuits against Anthropic, OpenAI, and others.

Legally, the case is straightforward for bear as I explicitly licensed it. The other two are a gray area of my own making. I meant to license them, and didn’t.

And how do I feel about it? Fine, honestly. First thought was cool! I gave this code away so other developers could learn from it and write better code. A model that learned from it and helps someone write better code is the same idea, just at a scale I never imagined.

The twist

This year I’ve worked on projects migrating off aging codebases, ColdFusion among them, using Claude Code to do most of the work. So code I wrote to help the ColdFusion community write better ColdFusion may end up helping companies migrate off of ColdFusion. It is also valid that it will help keep codebases alive supported in part by LLMs.

That’s open source. You give it away, and you don’t get to choose what it becomes.