Debian Votes on the Fate of LLMs in the Project

Debian has opened the discussion period for a General Resolution (GR) on the use of large language models (LLMs) and other generative AI tools in work on the project. The debate began on July 24, and it is set to culminate in a vote that will establish the community's official position.

Two options

Developers are essentially being asked to choose between two opposing approaches.

The first option is to ban outright any material created with the help or assistance of LLMs and other generative tools. The ban would apply broadly: to Debian source packages, all of the project's software, web resources, documentation and translations, as well as other official communication on behalf of Debian. In other words, AI would be left no room in either code or text.

The second option is to permit AI-assisted contributions, but to hedge them with a set of conditions. Such material would have to meet requirements for the legal compatibility of the tools used, the rules on licensing and attribution, ensure accountability and disclosure of the fact that AI was used, and a number of other conditions. Here AI is allowed, but under supervision and with a mandatory “acknowledgment” of its involvement.

Context

The dispute is not arising in a vacuum. Free software communities are increasingly confronting the consequences of the mass use of language models — from questions about the provenance and license cleanliness of generated code to a flood of automatically composed reports. The GNOME project, for instance, recently revised its vulnerability disclosure policy precisely because of an influx of AI-generated security reports. Debian, one of the oldest and most influential distributions, is now trying to work out its own ground rules before the problem sorts itself out on its own.

You can review the wording of the proposals and follow the course of the vote on the Debian voting page. The outcome of this GR will be of interest far beyond the distribution itself: Debian's stance often becomes a reference point for other projects weighing how to treat generative tools in development.

Source: Phoronix