Published September 21, 2026 · Added September 21, 2026

From blanket bans to AGENTS.md: open source tries to govern AI code

Freenode News surveys how open-source projects are replacing blanket AI prohibitions with explicit rules, citing a kernel selftest driver series that Jason Gunthorpe said was coded by Claude Code in about four days and then de-slopped, QEMU's RFC by Paolo Bonzini that reverses its prior refusal and pairs human docs with AGENTS.md, a PyTorch pull-request hook that can echo attacker-controlled filenames into model context as a prompt-injection channel, and an emacs-devel fight over a Hermes front end that critics say funnels users toward non-free network services. The through-line is that 'no AI' held because it was simple rather than enforceable, while licensing provenance, reviewer bandwidth, and security boundaries remain unresolved.

Freenode News surveys how open-source projects are replacing blanket AI prohibitions with explicit rules, citing a kernel selftest driver series that Jason Gunthorpe said was coded by Claude Code in about four days and then de-slopped, QEMU’s RFC by Paolo Bonzini that reverses its prior refusal and pairs human docs with AGENTS.md, a PyTorch pull-request hook that can echo attacker-controlled filenames into model context as a prompt-injection channel, and an emacs-devel fight over a Hermes front end that critics say funnels users toward non-free network services. The through-line is that ‘no AI’ held because it was simple rather than enforceable, while licensing provenance, reviewer bandwidth, and security boundaries remain unresolved.

Read the original story.

Source: Freenode