Latent Diagnostic Taxonomy for Prompt Injection Detection

This framework builds dimensionality-optimized classifiers to detect prompt injections and uses a diagnostic taxonomy to identify trustworthy decisions. By locating latent support vectors, it flags prompts susceptible to heuristic bias or attack magnitudes.