Open Weights Aren't Open Source — The AI Industry's Mislabeling Problem
Source: The Register: Biting the hand that feeds
The AI industry has weaponized "open" to describe model weights lacking source code, training data, and modification rights central to open source software. This creates a marketing category that grants users access without agency. Companies like Meta and Mistral claim openness while controlling model training pipelines, data provenance, and commercial redistribution—selling the appearance of transparency while gatekeeping the infrastructure that determines what these systems learn. As open weights proliferate, the semantic collapse between "downloadable" and "open" erodes actual open source norms and gives enterprises cover to shape AI development without the accountability genuine openness requires.