AI-assisted content and AI use

Version 1.0 · Effective 3 November 2026 · YK Global FZC LLC

We use AI tooling in our own work, and we think a company that advises on AI governance should be able to describe its own. This page covers both: how AI is involved in what we publish, and how we govern AI in what we deliver.

1. What AI does in our publishing

AI tooling assists with:

• research and source gathering,
• first drafts and structural suggestions,
• adapting an approved piece to a channel’s format,
• consistency, terminology and accessibility checks,
• draft translation, before human review.

2. What AI does not do

It does not approve, publish, or send. There is no autonomous publishing path and no autonomous external communication: our agents draft, and a person sends. Approval requires a named human editor, is bound to the exact version reviewed, and requires a quality score of at least 90 of 100 - described in our editorial policy.

It also does not decide what we claim we can do. Our record of assessed capability is maintained by people, and an automated check prevents an article from claiming more than that record supports.

3. A person remains accountable

AI assistance does not transfer or dilute responsibility. For every published item a named editor is accountable for accuracy, originality, judgement, tone, cultural fit and the decision to publish. “The tool wrote it” is not a position we will take, and we do not consider it an available one.

4. Translation is labelled, not laundered

This site is published in eight languages. Where a page carries source text that a speaker of that language has not yet reviewed, the page says so in the reader’s own language and is excluded from search indexing until it has been reviewed. We do not present unreviewed machine translation of a technical or regulatory claim as finished copy.

5. How we route work between models and people

Tasks are routed by what they require rather than by what is cheapest. Deterministic work - schema validation, link checking, cost arithmetic - is done in code, not by a model, because a model is the wrong tool for a question with one correct answer. Classification, extraction and drafting may go to a model. Strategy, publication and anything carrying regulatory or commercial commitment goes to a person.

Routing is policy that we can show you, separate from the code that calls any provider, so it can be reviewed and tested without a live model connection.

6. Confidential and client data

Content marked restricted is not sent to an external model. We do not use client data to train models, and we do not permit it to be used for training by a provider. Where an engagement requires it, work runs against a private or sovereign deployment; where that is not in place, the work does not use an external model.

7. We account for what we spend and what it produces

Model usage is recorded with token counts, provider charges and the human minutes spent reviewing the output. Costs are calculated exactly rather than with floating-point arithmetic, and an unknown price is never recorded as zero. We would rather report a gap than a figure that flatters us.

8. In client delivery

The same principles apply to systems we build for clients: human approval for consequential actions, permissions and audit trails on agent tooling, an evaluation set agreed before launch rather than after, and a documented model inventory. Where we deliver an AI capability with a partner, or where a capability is at research stage, we say so rather than implying we deliver it today - see our services.

9. What we will not do

• Publish AI-generated content as a person’s own unreviewed work.
• Generate fake reviews, testimonials, endorsements or engagement.
• Post automated comments or messages that present as a person.
• Use undisclosed brand accounts or manipulate votes on any platform.
• Send client or confidential data to a model that has not been approved for it.
• Claim a capability on the basis that a model could probably do it.

10. Questions and contact

If you want to know how a specific article or deliverable was produced, ask and we will tell you: info@ykglobal.ae.