· Updated
s2orc-safety on Hugging Face, a specialized slice of the S2ORC (Semantic Scholar Open Research Corpus) containing 16,806 academic papers. The collection focuses on critical safety domains including jailbreaks, prompt injection, red teaming, model security, privacy, robustness, and alignment.While academic safety research is abundant, it is often difficult to aggregate and compare across different studies. This release enriches each paper with structured fields for safety taxonomy, experimental details, and normalized names for models, datasets, and metrics. This standardization allows teams to programmatically analyze safety trends and reproducibility across thousands of documents.
You can access the dataset on Hugging Face to integrate safety research into automated evaluation pipelines or literature reviews. The inclusion of code-link metadata and practicality scores helps engineers identify which safety mitigations have functional implementations ready for testing. The repository was recently updated to ensure the most accurate metadata is available.
We're releasing 's2orc-safety' on @huggingface: a AI safety slice of our s2orc-enriched dataset with 16,806 papers across jailbreaks, prompt injection, red teaming, model security, privacy, robustness, alignment, and more. Each paper is enriched with structured fields for reproducibility, safety taxonomy, experimental details, practicality, normalized model/dataset/metric names, code-link metadata, and more. Link below:
More like this
OpenRouterApr 30
Nicolas KrassasApr 6
AnthropicApr 19