Register Owner: CTO (AIMS Owner) + General Counsel
Effective Date: June 13, 2026
Reviewed: On material AI change and at least annually
Frameworks: ISO/IEC 42001 (AIMS; supports the AIMS SoA); EU AI Act (Art. 10 data governance, Annex IV); supports the AI Model Registry.
How candidate/customer data reaches AI models, the deidentification applied before any training, the no-training-on-customer-data boundary at inference, and the human-oversight controls.

​
AI data flow

​
Controls on the flow

  • Inference (left/main path): customer data sent to providers transits Portkey; providers operate under enterprise terms that prohibit training on Neuroscale’s data. No raw customer data is retained by providers for training.
  • Training (boxed path): only deidentified data (per the Deidentification Standard) enters a training corpus; differential-privacy parameters apply where used. Re-identification risk is checked via the Re-identification Audit.
  • Human oversight: AI outputs informing consequential employment decisions are subject to human review; bias auditing per the Employment-AI Bias-Audit Procedure.
  • Transparency: AI-interaction and AI-content disclosures per AI Acceptable Use (EU AI Act Art. 50).
  • Registry: every feature’s model, tier, and DPIA/FRIA are recorded in the AI Model Registry.

​
Cross-references

​
Version history

VersionDateDescriptionAuthorApproved by
1.0June 13, 2026Initial AI model data-flow diagram.Cameron WolfeIshan Jadhwani