A water utility proposes feeding SCADA historian data into a generative training pipeline without OT security review. What lifecycle gate should AI security enforce?

When a water utility proposes feeding SCADA historian data into a generative training pipeline without OT security review, AI security must mark that historian data out-of-bounds for generative training until OT security authorizes a controlled path. SCADA telemetry can reveal process setpoints, network topology hints, and operational patterns that increase cyber-physical risk if absorbed into generative systems or vendor training environments. De-identification and existing IT data governance approval do not authorize a new processing purpose or OT risk acceptance. Permitting training on de-identified historian aggregates under IT governance workflows confuses data-governance sign-off with OT security authorization and purpose limitation. Authorizing vendor-hosted fine-tuning after a standard IT third-party risk assessment substitutes IT procurement process for OT security review. Forwarding historian extracts through DLP screening and detailed access logging before approving a generative fine-tuning pilot still routes OT telemetry into generative systems without OT security review. Exam caveat: even aggregated or sampled historian extracts may need OT review when they could reconstruct sensitive process behavior. Operational check: enforce a data-use gate that blocks generative training jobs from OT historian sources unless an OT security ticket documents approved subset, isolation environment, retention, and vendor non-training clauses.