Hi all, I'd like to add three questions to the podling proposal template, alongside the existing cryptography question:
1. "Does the project distribute trained model weights (as release artifacts, convenience binaries, or via package repositories)? If yes, briefly describe what models, where they came from, and under what license." 2. "Does the project's software download, depend on, or call third-party AI models or model APIs at runtime?" 3. "Has AI tooling been used to generate substantial portions of the codebase being donated?" Why: model weights are an artefact type our release guides handle awkwardly. LICENSE and NOTICE need to account for them, and provenance of training data raises licensing questions that code doesn't. OpenNLP is a good example for handling this well, with formal model releases that undergo the same voting process as code. The second question covers the increasingly common case where a project doesn't ship weights but depends on them at runtime. The third is a provenance question the ASF's Generative Tooling guidance covers, and knowing the answer up front makes IP clearance smoother for everyone. To be clear about what this is and isn't: - These are disclosure questions, not acceptance criteria. A yes to any of them does not affect the proposal vote. - A yes means mentors know what to look at during incubation: model provenance and dependencies at release time, and what applies from the Generative Tooling guidance. Nothing more. - Many projects will answer no to all three in a word each, but a yes to the third is entirely normal in 2026. The point is knowing, not judging. These questions don't create any new obligation. The existing release policy already applies to anything in a release, including model weights. For AI-generated contributions, the Generative Tooling guidance already applies. Recent regulation (the EU AI Act, among others) is the background context here. Unless there are objections, I'll make the edit to the template on the wiki in a week or so. Thanks, Justin --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
