transformers

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill explicitly loads public models and datasets and fetches external media (e.g., datasets.load_dataset("imdb", "conll2003", "squad", etc.), AutoModel.from_pretrained/snapshot_download and Hugging Face Model/Datasets Hub links, and the example image fetched from http://images.cocodataset.org), so it ingests untrusted, user-contributed third-party content that the agent is expected to read/interpret during training and inference.