EU AI Act impacts open-source GPAI models — guidance for Hugging Face, Gradio, and disclosures

AI Impact Summary

The EU AI Act is in force with risk-based obligations that apply to open-source AI systems, including general-purpose models (GPAI). The guide emphasizes practical tooling from Hugging Face (model cards, dataset cards, watermarking, opt-out and data redaction) and Gradio to help comply, signaling that release pipelines must embed metadata and user-facing disclosures. The risk framework references a systemic threshold (10^25 FLOPs) and specific high-profile models (Gemini 1.0 Ultra, Llama 3.1-405B, GPT-4, Mistral Large, Nemotron-4 340B, MegaScale, Inflection-2, Inflection-2.5) to illustrate where stricter obligations apply, while most open-source GPAI projects will land in the limited-risk category with transparency requirements. Enforcement starts August 2026, with a two-year transition; teams should audit data sources, model provenance, documentation, and user-facing disclosures for EU-relevant deployments now to avoid disruption.