Enterprises face risk from AI training on incident archives

Security incident records are increasingly being ingested by security AI systems to train and validate models, creating new risks for enterprises regarding data exposure and legal accountability. The author argues that ordinary security records acquire a sensitive second use when vendors use them for model improvement, potentially exposing patterns of detections and tuning decisions. The article contends that evidence retention and model development should be formally separated in procurement contracts requiring explicit, revocable customer choice. Open article →