AI assistants and the open-source maintainer's inbox

The new failure mode

An AI-drafted issue that looks well-formed but describes a behavior that doesn't exist. A PR that compiles but breaks an invariant the maintainer cares about. The polish makes the noise harder to filter than the old 'one-line bug report' problem ever was.

What the healthy projects are doing

• Requiring a minimal reproduction in a sandboxed repo, not pasted code.
• Auto-closing issues that match templates of known-AI hallucinations.
• Asking PR authors to confirm they've run the test suite locally.
• Publishing a short 'how to contribute with an AI assistant' guide.

What contributors should do

Read the contributing guide. Run the tests. Don't open the PR if you can't explain what it does in your own words.

Above all, don't open three PRs hoping one will land. Pick one, do it well.

Where this is heading

Expect more projects to require signed CLAs, sandboxed reproductions, and proof of test runs. Expect a handful of large projects to ban AI-generated contributions outright, and others to embrace them with strong gating.

A note for users

If an AI assistant helped you find a bug, that's great. The contribution still needs to meet the project's bar. Treat the maintainer's time the way you'd want yours treated.