TRUST CENTER

AI and automated processing

How AI features are bounded and where human review remains required.

Last reviewed: 31 August 2026

Implemented

Workspace choice

A deploy-wide kill switch and audited owner/admin workspace controls can disable all AI, Knowledge Intelligence AI, or behavioural-intelligence AI. Enforcement occurs on server service boundaries.

Implemented

Bounded requests

AI providers receive only the context assembled for the requested feature. Provider response storage is disabled in Beetl integrations, output sizes and timeouts are bounded, and untrusted content is separated from instructions.

Implemented

Human review

Generated content is a draft or recommendation. Publication and governed changes remain separate, permission-checked actions.

Recorded

Traceability

Feature records retain provider/model, prompt or schema versions, source-version identifiers, token counts where returned, timestamps and outcomes. Assistant questions are retained under the workspace AI-interaction retention period; generated answers are not stored in the assistant interaction record.

Contract verification required

Provider commitments

Provider training, abuse-monitoring exceptions, processing location and deletion commitments must be verified against the current provider agreement before customer data is enabled.