Trust center · AI data use

AI may assist work; it does not silently acquire authority.

WythinForge governs AI by data boundary, evidence, approved purpose and human control—not by a generic claim that AI is used.

Last updated
25 August 2026
Status
Current Workspace boundary and product policy

Current Workspace boundary

The Workspace MVP has no customer-facing AI assistant and does not send customer industrial transactions to a model. Deterministic services remain the production default for request and delivery state transitions.

Rules for AI-assisted processing

When an approved product workflow uses an AI model, prompts, schemas, model configuration and evaluation evidence are treated as versioned delivery inputs.

  • Customer data is not used by WythinForge to train a shared foundation model.
  • A provider and data path must be approved for the stated purpose before customer content is transmitted.
  • Model output is untrusted input: deterministic validation and authorization still apply.
  • High-impact actions remain proposal-only until an authorized human approves them.
  • Human corrections become governed change proposals and must pass regression evaluation before a production rule or prompt version changes.

Evidence-first product direction

The planned Inbound AI Handler will bind extracted values to source evidence, confidence, deterministic rules and human review. It will be developed in a separate repository; this Workspace remains the commercial and operational control plane.