Many people believe open source is shutting the door on AI written code. I read two studies from this year that checked this properly and tried to validate with understanding from open source projects I am involved in.
One study looked at the 2000 most popular projects on GitHub and added 36 well known ones like the Linux kernel and Apache. It found 281 written AI policies and read every one of them by hand. The other study read the rules of 67 highly visible projects along with the arguments and incidents behind them.
Lets start with how fast the rules are appearing.
a) Ai usage policy files barely existed before this year. Of the 92 dedicated policy files the first study tracked only 8 were written in 2025 and 84 were written in 2026.
b) Half of them have already been changed at least once and they get longer each time. The changes almost always add stricter quality rules and clearer disclosure and more human responsibility.
c) The overall direction is still toward allowing AI. The same authors found 78% of policies were permissive in an earlier study and they found 83.3% a few months later.
d) Every project is writing its own version. There is no shared template and no common wording. This I feel is ripe for standardization.
Lets look at what the policies allow.
a) 83.3% of the policies allow or encourage AI in code contributions and 14.9% ban it.
b) These permission comes with rules. 67.3% of the policies say a human must stay closely involved.
c) 27.8% of the policies ban AI in comments and discussions while only 13.9% allow it.
The second study found that projects fall into three groups
a) The first group treats AI written code as something that should not enter the project at all and the reason is usually legal and not quality.
b) The second group lets AI code in but only with disclosure and human accountability and proof that it works.
c) The third group does not write any AI-specific rules and instead applies its existing quality standards to everyone.
Lets look at whats required of committers.
a) 43.4% of the policies say the human who submits the work is responsible for all of it including the parts the model wrote.
b) 48.8% of the policies ask you to say that you used AI.
Note to committers - If you contribute to several projects you can follow one set of rules and break another without knowing it.
Lets look at what projects do when the work is AI slop
a) The three most common are closing the pull request without much review at 48.4% and banning the person at 20.8% and refusing contributions from fully autonomous agents at 13.8%.
b) Some projects limit how many pull requests one person can have open. Some keep a public list of people who send bad AI work.
Overall writing code has gotten cheaper and reviewing it has become more expensive. So projects are moving the checks like who is accountable for change and proof that change works before review can begin.