NEW YORK — What happened: Major artificial intelligence laboratories have updated the licensing agreements attached to their latest open-weight model releases, introducing revenue thresholds and restrictive downstream usage clauses. While weights remain downloadable, the definition of "open" is diverging sharply from traditional open-source software standards.
Why it matters: For independent developers and small businesses, building products on top of open-weight models has been a cost-effective alternative to proprietary APIs. These new licensing structures introduce legal ambiguity, making long-term product planning risky for companies crossing specific financial milestones.
Deep dive: Traditional open-source licenses focus on permissions rather than user identity or commercial revenue. In contrast, modern open-weight licenses operate more like source-available or dual-licensing frameworks found in enterprise databases. Labs argue that these restrictions protect foundational investments and prevent misuse, while critics point out that conditional permissions stifle ecosystem growth and create uncertainty.
Report check (claims vs what is verified vs still rumor): Verified: Multiple foundational labs have published revised end-user license agreements with tiered commercial limits. Claimed: Some industry observers assert these licenses will eventually ban self-hosting entirely. Rumor: Discussions suggest future EU regulatory frameworks may legally mandate stricter definitions for open-source AI, though nothing is finalized.
Open questions: How will enforcement of revenue-based clauses actually work in practice, and will courts uphold restrictions on the deployment of model weights that have already been publicly downloaded?