Pinned Loading

1. agentshield agentshield Public

   Security audit framework for agentic AI systems: STRIDE threat modeling, 100-scenario attack suite, 4-component detection pipeline (96% detection, 1.0% FPR)

   Python

2. ambiguity-casebook ambiguity-casebook Public

   Dual-Use Ambiguity Casebook: 30 structured cases at the AI-era biology dual-use decision boundary

   Python

3. bio-constitution-rules bio-constitution-rules Public

   30 machine-readable constitutional rules for biological dual-use content across 6 bio domains. JSON format for Constitutional Classifier pipeline integration.

   Python

4. bio-overrefusal-v0.1 bio-overrefusal-v0.1 Public

   Domain-expert-authored benchmark for LLM over-refusal on legitimate biology research queries.

   Python

5. constitutional-bioguard constitutional-bioguard Public

   Biological dual-use content classifier using Constitutional Classifiers methodology — biosafety constitution, synthetic data pipeline, DeBERTa-v3-base classifier

   Python

6. narrow-model-safety-eval narrow-model-safety-eval Public

   Empirical dual-use risk assessment of protein language models (ESM-2) and structure-based design tools (ProteinMPNN)

   Python