Healthcare

Protected Health Information (PHI) Masking: Identify the representation associated with Protected Health Information (PHI) such as patient names or medical record numbers. Use RE to instantaneously mask or censor this representation within the output stream, enforcing data privacy and HIPAA compliance.

Harmful Intent Suppression: Isolate the vectors corresponding to self-harm, violent, or otherwise inappropriate content. Representation Steering is used to suppress the activation of these representations, diverting the model to provide safe, supportive, and compliant responses.

Non-Authoritative Advice Steering: Define the vector of "authoritative clinical guidance" versus "unsafe diagnostic speculation." RE acts as a guardrail, maximizing the authoritative representation while suppressing non-compliant or potentially harmful medical advice generation.