AI Disclosure and Accountability

Effective August 21, 2026. Applies to learnxiangqi.com.

1. Who is answerable

Learn Xiangqi is operated and published by SproutVest, LLC. Every fact, every line of code, every piece of coaching advice, and every consequence of getting one of them wrong is ours. "The AI wrote it" is not an excuse, and we will not offer it.

2. Where AI is involved in building this

The code. Most of the software behind Learn Xiangqi was drafted with AI assistance. It is reviewed, tested and gated exactly as hand-written code would be; nothing ships because a model produced it. The engine that plays you is ordinary search code, not a model.

The writing. Much of what you read here — the guides, the interface, the legal pages, and this page — was drafted with AI assistance and edited by a person before publication.

Translation. This site is published in English and Chinese. The translations are AI-assisted and checked for meaning by a person who is not a native speaker of both. That is a real limit, and it is the one place where your eyes are better than ours: if a page reads wrongly to you, tell us.

3. Where AI is involved in using this — the Analyze button

This is the part most products do not have, so it gets its own section rather than a deferral. Only when you press "Analyze", we send the following to Anthropic, PBC for processing by the Claude API:

• a text rendering of the current board position;
• the list of moves played so far in the current game;
• whose turn it is; and
• any question you typed into the panel.

We do not send your email address, your account identifier, or your IP address with that request. Through our commercial API access, Anthropic does not use the content of these requests to train its models. If you never press Analyze, nothing is ever sent. The Privacy Policy is the authority on this and section 3 there is the longer version.

What comes back is coaching advice, and it is exactly as fallible as section 4 says. It is not a verdict on the position, it is not a guarantee that a move is best, and a strong opponent will punish a plan it endorsed. Treat it as a second opinion from a good club player rather than an oracle.

4. Accuracy

AI systems produce fluent, confident prose that is sometimes wrong, and the fluency is the difficult part: a mistake made this way does not look like a mistake. So we check what we publish, and we do not claim the checking is perfect.

Our automated gates prove that a page exists, that it says the things it is required to say, and that a shared file has not silently changed underneath us. No check anywhere proves that a sentence is true — and no gate reads the coaching advice at all, because it is generated fresh for a position nobody has seen before.

When you tell us something is wrong, we will look at it, fix it if it is, and say plainly that we did. We will not quietly edit a page and leave the record implying it always said that.

5. Bias

Models are trained on what people have written, and they carry what people have written — about nationalities, genders, ages, and whose way of playing or speaking gets treated as the correct one. Xiangqi is a game with a deep Chinese literature and a much thinner English one, and a model trained mostly on English text will know the English framing better. Where the coach explains an idea in the vocabulary of international chess rather than of xiangqi, that is what you are seeing.

We treat a reported bias as a defect with a fix, not as a difference of opinion. And we do not use AI systems to score, rank, screen or judge a person.

6. Rights and provenance

AI systems are trained on work that other people made. There is a live legal and ethical argument about that which this page does not pretend to settle. What we will commit to is narrower and checkable:

• We do not prompt a model to reproduce a particular person's work or to imitate a named living creator's style.
• We do not knowingly publish material that is substantially copied from someone else's work, and we remove it when we find out.
• We do not generate a real person's likeness, voice or name to suggest an involvement or endorsement that does not exist.
• Third-party fonts, images and code carry their own licences, and we keep those licences with them.

If you believe something we publish infringes your work, write to [email protected] with enough detail to find it. We will review it and attribute, licence or remove it.

7. Your data and AI

Apart from Analyze, described above, nothing you do here is sent to an AI provider, and nothing you do here is used to train a model. Your play, your settings and your progress are not analysed by a model to decide anything about you.

We also do not use an AI system to make a decision about you that affects your money or your access — a purchase, a refund, a suspension — without a person making it.

8. What we will not do

• We will not publish AI-generated text as though it were a person's first-hand account. No invented reviews, testimonials, case studies, credentials, or people.
• We will not present AI output as professional advice. The coaching is instructional, and the Terms of Use say so.
• We will not use this page as a defence. Disclosing that AI was involved changes nothing about our responsibility for what it produced; it is a courtesy to you, not a limitation of liability.

This statement was itself drafted with AI assistance and read by a person before it was published, which is exactly the process it describes.

9. Telling us we got something wrong

Bad coaching advice, a translation that reads badly, a biased example, or a passage you think is someone else's: [email protected].

Legal notices: [email protected]. Privacy questions and data rights requests: [email protected].

10. Changes

If this statement changes materially, the effective date above changes and the revised version is published here before it takes effect.