Debian developers challenge unannounced LLM package reviews

Debian Developer Johannes Schauer Marin Rodrigues says he will stop uploading packages to the NEW queue after discovering that submissions undergo an unannounced large language model review workflow. The reports disclaim review by a DFSG team member and state that they are not package decisions. He argues that maintainers must either spend time checking potentially incorrect feedback or ignore output that still consumes computing resources.

Critics say the rollout conflicts with the spirit of Debian’s recent general resolution on generative AI. The resolution neither endorsed nor banned the tools, but called for prior discussion and consensus on large-scale automated actions. Marco d’Itri said the DFSG team should take responsibility for the feedback rather than ask maintainers to process unreviewed model output.

Mathias Gibbens reported reviews ranging from several hundred words to more than 1,500 and said the process appears to favor uploads that satisfy model complaints. Matthias Geiger identified incorrect advice about bundling an already packaged library and a false blocker involving a Build-Depends acceptable in experimental.

Shengjing Zhu noted that the messages can be filtered and human reviewers should retain final authority. Other developers highlighted an open opt-out request and argued that ignoring model feedback should carry no penalty unless a human reviewer validates it.