Hugging Face Hub response for meta-llama/Llama-Guard-3-1B: 'Access to model ... is restricted. You must have access to it and be authenticated'

Directly measured: an unauthenticated GET of the model config returns a gated-access refusal. This is the repository-side half of the model-access-security story - the licence and the weight host enforce access separately from any API key on your own inference endpoint. The API key protects your serving capacity; it never protected the weights.