AI Transparency Statement

1. Where we use AI

2. Where we do not use AI

No content is generated by AI when you open a page. The regulation summaries and guides, deadlines, use cases, templates, the glossary and the HKU Research Project results are prepared by the committee and checked against the official texts and the underlying sources before they are published. The Self-Assessment scores your answers with a fixed formula in your browser. Provider listings come from the member companies and need committee approval.

We use AI tools while building the platform, for example to help write code and draft texts. Everything that goes live is reviewed by a committee member first; items still under full committee review, such as the templates, are labelled as such. The figures and the robot on this platform are drawn as code; we do not use AI-generated images of people.

No AI output decides anything about you: not your access, membership, listing or any other outcome.

3. Model and provider

Both features use a Grok large language model from xAI, called through xAI’s API from our server functions in the United States. xAI may process a request in the United States or in another region. The model’s general knowledge has a cut-off date and does not know about recent regulatory changes, which is one reason it is not allowed to state regulatory facts itself (see below).

4. Safeguards

• Closed list of destinations. The assistant may only choose from a fixed list of this platform’s pages, plus the Chamber’s ESG Sourcing Hub. Any page it names that is not on the list is dropped before the answer reaches you, so it cannot send you to an invented or manipulated link.
• No regulatory facts from the model. The assistant is instructed never to state regulatory dates, deadlines, fines or legal requirements itself, and to point to the regulation guides and deadlines instead, where the committee maintains checked information.
• No invented providers. Matchmaking can only rank companies from the provider directory. Any company name the model returns that is not in the directory is discarded.
• Your text is read as a question. The model is instructed to treat what you type as a question or a need description only, never as instructions. Because of the closed list of pages and the directory mapping, even a manipulated answer cannot send you to an invented link or company.
• Short, bounded answers. Questions to the assistant are limited to 500 characters, matchmaking descriptions to 1,500, and answers to a few sentences. Hourly limits per visitor and per member sign-in protect the service from abuse.
• Graceful fallback. If the AI fails or takes too long, you get the keyword search or the category results instead of an error or a guess.

5. Your questions and data

Your question, or your matchmaking description, is sent once to xAI to produce the answer, together with our list of pages or the published profiles of the listed member companies (never contact persons or websites). We do not store it on our server and we do not use it to train any model. The conversation stays in your browser tab and is gone when you close the tab or sign out. xAI does not train its models on API requests; it may keep them for up to 30 days to investigate suspected abuse, then deletes them.

Please do not enter personal data about other people, or confidential or commercially sensitive information. The Privacy Policy has the details, including the hourly limits, which briefly keep your IP address or a one-way code derived from your sign-in token in memory.

6. Limits

AI answers can be incomplete or wrong, even with these safeguards. They are orientation, not legal or professional advice, and not the opinion of the Chamber or the committee. Check them against the pages they point to and, for regulation, against the official texts. The Terms of Use apply.

7. Oversight and feedback

The AI Committee is accountable for the AI features. It decides which pages and companies the AI may point to, sets the instructions the model works under, and reviews them when the platform changes. Because we do not store questions, we rely on you to tell us about a wrong, unhelpful or inappropriate answer: write to info@hongkong.ahk.de with the question you asked and the answer you got.

We update this statement whenever we add, change or remove an AI feature, or change the model provider.

8. Frameworks we follow

We apply the practices we recommend to members: the Hong Kong Privacy Commissioner’s Artificial Intelligence: Model Personal Data Protection Framework, and the transparency principle of the EU AI Act, which says people should know when they are interacting with an AI system. The Security & Governance pillar of our AI Blueprint describes the same approach for companies.