Responsible AI policy

Principles and controls for responsible AI development and deployment.

Byt-Wyze principles and controls for responsible AI development and deployment.

Byt-Wyze exists to make evidence, uncertainty and confidence measurable. A platform that makes that claim has to hold itself to the standard it sells. This charter states the principles we design and operate against, and how each is put into practice.

We prefer methods whose behaviour can be explained over methods that merely perform well. Model assumptions, elasticities and parameters are documented and, in the policy suite, exposed for audit. An unexplainable result is treated as a defect, not a feature.

Results are reproducible from their inputs. API responses record the parameters used, model versions are recorded, and calibration changes are logged so that a historical result can be reconstructed.

Our systems are decision support. They are designed to be interrupted, overridden and questioned. No Byt-Wyze product makes an automated decision producing legal or similarly significant effects on an individual without human involvement, and none is marketed for that purpose.

AI systems inherit the security posture of the infrastructure around them. Access to models and APIs is authenticated and rate-limited, secrets are held server-side, and our security practices are set out in the Security Policy.

We minimise personal data. Our analytical products operate on parameters, integers, aggregate statistics and customer-supplied artefacts rather than on personal data, and we do not use customer inputs to train models.

  • Responsible AI Charter
  • We state the data sources and assumptions behind each model, including their geographic and demographic scope.
  • We avoid extrapolating a model calibrated on one population to another without flagging it.
  • Where distributional effects matter (as in policy modelling) we report them rather than reporting only aggregates.
  • We do not build or supply models for social scoring, biometric categorisation or emotion inference.