Trust & transparency
How Trai scopes AI systems, protects business data and keeps people in control of important decisions.
Last updated: 10 October 2026. This policy sets the delivery boundaries for AI projects, not a claim that every advertised AI feature runs on this website. The current enquiry form sends an enquiry to Trai; it is not an AI advice service. Project-specific tools, purposes, approved data and responsibilities must be agreed before implementation.
Customer-facing voice or chat agents must identify themselves as AI and offer a practical human handover. Urgent, sensitive, disputed or out-of-scope requests need an agreed fallback. Financial commitments, scope changes, refunds and other consequential actions require appropriate human approval. AI must not make autonomous high-impact decisions or provide regulated professional advice.
Use only information the client is authorised to provide, and minimise personal or sensitive data. Before connecting a model or agent, agree permitted sources, access permissions, data locations, retention, deletion and provider terms. Do not assume a consumer AI account offers the controls needed for business data. Client or customer data must not be used to train models without contractual authority and specialist review.
AI output can be wrong, incomplete or misleading. Test realistic tasks, ambiguous questions, failed integrations and attempts to override instructions. Define what the system may answer, what it must refuse or hand over, and how approved knowledge is updated. Logs should be access-controlled, limited to what is needed and subject to agreed retention. Monitoring, rollback and incident escalation must be part of the scoped operating plan.
Recording, monitoring and transcription require appropriate notice or consent based on the applicable jurisdiction and use. Commercial email, SMS and outbound calling need appropriate consent and suppression controls; do not use harvested lists. Any legal, tax, safety or regulated document content must come from the client’s appropriate qualified adviser. Technology implementation does not authorise Trai to provide that advice.
If an AI-assisted interaction in a Trai-implemented project seems wrong, contact the business operating it and request a person. For questions about Trai’s own handling of data, contact Trai and identify the relevant system or interaction without sending sensitive records in the initial enquiry.