Results for “reaction-mechanisms”

Probe RAG applications for prompt injection via poisoned retrieved context and embedding manipulation.

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.