1. Summary
1.1AI on this platform only drafts, summarises and suggests. It never approves, rejects, verifies, ranks, licenses, sanctions or moderates anyone. Every decision about an athlete, a result, a licence or a member is taken by a person.
1.2AI features use the Claude models of Anthropic PBC, called from our servers through Anthropic's API. Your browser never talks to Anthropic directly. AI features are currently switched off for this site; where a feature is shown, it uses a fixed, non-AI template instead.
1.3AI features can be switched off individually by KBN, and are hidden when no API key is configured.
↑ Contents2. Which features use AI and what is sent
| Feature | Who uses it | What is sent to Anthropic | What happens to the output |
|---|---|---|---|
| Scouting report | Anyone, on a fighter profile (on click) | The fighter's public, verified record: name, country, division, verified results, methods and points | Shown as an informal three-paragraph summary; cached until the record changes |
| Registration assistant | Applicants on /apply | Your question and KBN's published ranking rules. Do not type personal data into it | An answer based only on the rules; where the rules are silent it says the Ranking Committee decides |
| Document pre-check | Branch and HQ administrators | The list of documents attached to an application and the claimed bouts (not the files themselves) | Saved as an application note labelled "AI pre-check"; advisory only |
| Document reading (OCR & auto-fill) | Applicants and administrators, where enabled | The uploaded document image or PDF, which may be an identity document or result sheet | Suggested field values for a person to check and confirm; nothing is saved to the record automatically |
| Result verification assistant | Administrators | The claimed result and the evidence description | A draft note of points to check; the administrator decides |
| Matchmaker | Event administrators | Public ranking and record data of candidate fighters | Suggested pairings; the organiser decides |
| Live commentator | Live centre | Round scores and public bout information | Short round summaries labelled as AI-generated |
| Oracle and premium analysis | Fans | Public ranking and record data for the two fighters in a bout | An informal prediction for entertainment, not betting advice |
| Newsroom drafts and quiz questions | Editors | Public source articles, results and editorial notes | Drafts only; an editor must review and publish |
| Admin assistant | Administrators | The question asked and the relevant extracts of KBN rules and audit-log entries needed to answer it | An answer for the administrator; it cannot change anything |
3. What AI is never used for
- Deciding whether a result is verified, an application approved or rejected, or a licence issued, suspended or revoked.
- Calculating ranking points (the ranking engine is a fixed formula, see Ranking regulations).
- Moderating, suspending or banning KBN Social members, or deciding verification badges.
- Making any other decision with a legal or similarly significant effect on a person.
- Advertising or profiling.
4. Human review and accuracy
4.1AI output can be wrong, incomplete or out of date. It is always labelled or presented as a draft, suggestion or informal summary. The person using it is responsible for checking it before relying on it.
4.2If you think AI-generated content about you is inaccurate, tell us at holdingkbn@gmail.com and we will review and correct or remove it.
↑ Contents5. How Anthropic handles the data
5.1Anthropic acts as our processor under its commercial terms and data processing addendum. Under those terms, data sent through the API is not used to train Anthropic's models, and is retained by Anthropic only for a limited period for trust and safety purposes. The exact retention period and transfer mechanism are [to be confirmed] against the current Anthropic agreement.
5.2Anthropic is based in the United States; see the international transfers section of the Privacy notice.
↑ Contents6. Logs and limits
6.1For each AI request we record the feature, the user (if logged in), the model and the number of tokens used, to control cost and prevent abuse. We do not log the full prompt in that usage log. AI features are rate-limited per user.
6.2The model currently configured is claude-sonnet-5.
↑ ContentsQuestions about this document: holdingkbn@gmail.com. Where this document is translated, the English version prevails to the extent permitted by law. Version 2.0, last updated 28 September 2026. Printed copies may be out of date; the version at worldkbn.com/legal/ai applies.