Client choice
Projects can prohibit AI, permit specified AI uses, or require an approved enterprise/private setup. The setting should be visible in the project brief.
AI GOVERNANCE
Client-controlled AI, not silent AI.
Projects can disable AI entirely, restrict processing to private enterprise models, use approved providers, or use a client-provided API key.
Professional standard
AI is a processing choice, not a hidden assumption
AI can help with intake, classification, terminology work, drafting, translation, QA and multilingual analysis. Whether it should be used depends on the content, confidentiality, risk, client instructions and the approved technical environment.
Data sensitivity
Passports, medical records, contracts, unpublished products and regulated information need stricter handling than public marketing copy.
Human accountability
AI output is not treated as evidence of correctness. For professional delivery, a qualified person remains responsible for the level of review promised to the client.
Model boundaries
General-purpose models can hallucinate, omit text, normalize names, alter numbers or invent terminology. Those failure modes matter especially in legal, medical and technical work.
Approved language assets
Terminology, translation memory and reference material should constrain the workflow where consistency matters rather than allowing each AI call to improvise.
Traceability
Where AI materially affects production, the organization should be able to explain the approved use and the human controls applied afterward.
Human-governed AI
A simple risk rule
The more consequential the content, the less acceptable it is to rely on an opaque output without professional verification.
For low-risk, high-volume content, automation may provide useful speed. For customer-facing, legal, medical, financial, safety-critical or reputationally sensitive content, review depth rises because the cost of a plausible-looking error is higher.
This policy is designed to preserve client choice and professional responsibility. It is not a promise that every possible AI provider or deployment model is available in every production environment.