Rocky Linux founder launches OpenWALDO to bring transparency to AI training data

The AI industry has long struggled with opacity around training data sources and licenses. The MIT-led Data Provenance Initiative's audit of over 1,800 text datasets revealed a systemic problem: license information was omitted in over 70% of cases and miscategorized in more than 50%. This matters because models trained on data with unclear or incorrect licensing create legal and ethical risks for organizations that deploy them, and because training on outputs from other AI systems without proper attribution can lead to model collapse—a degradation in model quality caused by recursive training on synthetic data.

Kurtzer's proposal frames OpenWALDO as the missing layer in the "open AI" movement. While open-weights models (those whose parameters are publicly available) and open-source AI code have become more common, the training data itself has remained largely proprietary and unauditable. OpenWALDO introduces an AI Bill of Materials (ABOM) that tracks data lineage, licensing, and sources through training and release—practices borrowed from software development. The project positions this as shared infrastructure that would allow organizations to spend resources on differentiation rather than repeatedly sourcing and validating basic training datasets from scratch. Kurtzer explicitly draws a parallel to Linux's success, arguing that inspectability, forkability, and community validation—not certification—drove adoption.

The real test will be adoption. OpenWALDO must attract sustained contributions of high-quality, legally usable data; establish governance processes that command trust; and demonstrate that provenance records provide practical value in model-development workflows. CIQ will provide commercial support but not control the project, a structure intended to preserve its community character.