A quarter of pull requests to open source projects are AI slop

A tool that writes code faster can still make the project it is written for worse. A Linux Foundation survey released this week and reported by Richard Speed at The Register puts a number on the problem: open source maintainers estimate that a quarter of the pull requests landing in their projects now are low-quality AI output, and on larger projects the figure is 31 per cent.

The Register reports that respondents to the Foundation's research estimated AI cuts code-authoring time by 52 per cent on average, and that 71 per cent of maintainers think generative AI has lowered the barriers to contributing to open source. Only just over half, 53 per cent, think the technology has improved community trust. For highly complex tasks, 21 per cent of respondents report AI took longer than writing it by hand would have done.

Where the overhead has moved

The headline finding, as the Foundation frames it, is governance. Only 35 per cent of surveyed projects have a formal policy permitting generative AI use, 32 per cent have no policy and no plans for one, and the rest are somewhere in between. Hilary Carter, the Foundation's SVP of Research, told The Register: "Open source projects need to design for a growing AI-generated contributor base, not simply react to it. That means setting clear expectations for GenAI, putting responsibility for testing and disclosure on contributors and giving maintainers better tools to manage incoming pull requests."

The curl project is the pattern. On the same day, maintainer Daniel Stenberg announced curl will move low and medium severity flaws to public view, because the security team is "struggling as a project as all key maintainers feel strained, overloaded and brought to the brink of what we can muster." That change is one way to adapt to a mail folder full of AI-assisted reports a small team cannot triage one by one.

The number that is missing

The Linux Foundation survey measures estimated time saved by contributors, and estimated share of slop seen by maintainers. It does not measure the hours a maintainer now spends triaging the slop, or the compile and test minutes the slop costs CI before it is rejected. Those are the costs that a 52 per cent authoring saving has to clear before the net is positive, and the Foundation's own reporting acknowledges that half of maintainers do not see the trust side of the balance improving.

For a team shipping AI-written patches to projects they do not maintain, two rules fall out of this number. Disclose when the diff is agent-written, so the maintainer can route it through the AI slop lane rather than the human one. And run the project's own lint, test and CI locally before the pull request opens, because every failing CI run on a stranger's repository raises the maintainer's slop estimate by one.

Source

AI cuts coding time, but it's also filling open source's inbox with crap, Richard Speed, The Register, citing a Linux Foundation maintainer survey.