AI Use

YachtSync uses artificial intelligence to read your certificates and extract information from them — titles, dates, certificate types — so you don’t have to type them in by hand. This page explains exactly what gets sent to AI, what doesn’t, and what happens to it. We don’t marketing-fluff this. If we’re sending your data somewhere, you should know.

Questions? legal@yachtsync.uk.

Summary of Key Points

The honest TL;DR.

1. Which AI we use

YachtSync uses Anthropic’s Claude, accessed via their commercial API. The specific model is claude-haiku-4-5 — a vision-capable model that can read both photographs of certificates and PDF documents.

The AI is hosted by Anthropic. We don’t run our own AI models, and we don’t fine-tune the underlying model. Each scan is a fresh API call to Anthropic’s standard infrastructure.

2. What gets sent to AI

When you tap “Scan” on a certificate, the App sends Anthropic three things:

1. The document itself — the photo or PDF you’ve just captured
2. A prompt — an instruction asking Claude to identify the certificate type and extract the title, issue date, expiry date and issuing authority
3. A small set of anonymised correction examples — to help the model do a better job (see Section 8)

That’s the entire payload.

3. What does NOT get sent to AI

Equally important is what we never send to Anthropic when you scan a certificate:

• Your name, email or password
• Your profile information (rank, vessel, training location, career goal)
• Other certificates in your account
• Your account ID, device ID or any tracking identifier
• Location data of any kind
• Anything from other users

The Claude API call is intentionally minimal. Anthropic doesn’t know who you are, which yacht you’re on, or even that you’re using YachtSync — only that an authenticated request came in with a document.

4. What Anthropic does with your data

Under Anthropic’s Commercial Terms of Service, which govern our use of their API:

• They do not use your data to train their models. This is a commitment in their commercial agreement — not a vague marketing claim.
• They retain API request data for up to 30 days, solely for trust and safety monitoring (e.g. detecting abuse of the API). After 30 days, the data is deleted.
• They don’t share your data with third parties for marketing or any unrelated purpose.
• They don’t sell your data.

For the canonical, current version of these terms, see Anthropic’s Privacy Policy and Commercial Terms.

5. Where AI processing happens

Anthropic’s API infrastructure is primarily located in the United States. When you trigger an AI scan, your document is briefly transmitted to the US for processing, the result is returned, and the document is held by Anthropic for up to 30 days as described above.

This is an international data transfer for UK and EU users. We rely on the UK International Data Transfer Agreement or the UK Addendum to the EU Standard Contractual Clauses as the legal mechanism for this transfer. These are the safeguards approved by the UK Information Commissioner’s Office.

If you’re uncomfortable with US data processing, you can simply choose not to use AI scanning — see Section 6.

6. AI is optional

AI scanning is a convenience feature, not a requirement. You can use YachtSync entirely without it.

When uploading a certificate:

• Tap Scan → AI extracts the details
• Tap Manual entry → you type the details yourself, no AI involved, nothing leaves the EU

If you have a strong preference against AI processing for any reason — client confidentiality, organisational policy, personal preference — manual entry is fully supported and works identically except for the typing.

7. Accuracy and your responsibility

AI is impressive but not infallible. Here’s what you need to know:

• Mistakes happen. Claude can misread dates, mistake one certificate type for another, or fail entirely on poor-quality scans. You will see incorrect extractions occasionally.
• You always review. The App shows you what Claude extracted before saving. Nothing is committed to your account until you confirm or correct it.
• You are responsible for the correctness of your records. YachtSync helps you organise your certificates — it does not replace your professional duty to know what’s valid and what’s expiring.
• YachtSync is not a certifying authority. The AI doesn’t verify whether a certificate is genuine, accepted by a particular flag state, or sufficient for a particular role. That’s outside its scope.

If you spot an AI extraction mistake, use the “Provide feedback” option to flag it. This helps the system get better — see the next section.

8. How AI learns from feedback

YachtSync gets smarter over time by learning from user corrections.

Here’s how it works:

1. When you correct an AI extraction, the correction is stored in our feedback database
2. Before each new AI scan, we fetch a sample of recent corrections and include them as “learning examples” in the prompt sent to Claude
3. Claude uses these examples in-context to improve its accuracy on the new scan

This is sometimes called retrieval-augmented learning or few-shot prompting. The Claude model itself is not modified or fine-tuned — we just give it useful context.

Privacy of feedback

We’re careful with this. Before any feedback is included in another user’s scan prompt:

• Personal identifiers (names, document numbers, passport numbers, dates of birth) are filtered out
• Only the corrected pattern is kept — e.g. “the AI mistook this for a generic Travel Document but it was actually a Passport”
• Your individual corrections are never shown to other users in any user-facing way

This means the system gets smarter for everyone, but no individual user’s document content is exposed to anyone else.

9. Why we chose Claude specifically

We evaluated several AI providers before choosing Claude. The reasons we picked Anthropic:

• Strong commercial terms. Anthropic’s API does not train on customer data — this is in writing, not just a marketing claim.
• Vision quality. Claude is one of the best models available for reading documents — particularly handwritten or photographed certificates.
• Maritime knowledge. Claude knows enough about STCW, MCA and flag state requirements to make sensible inferences (e.g. that an STCW BST renewal is typically 5 years).
• UK-friendly governance. Anthropic publishes meaningful safety and policy documentation; we’re comfortable with their corporate posture and data handling.
• Reliability. The API has high uptime and predictable response times.

If we ever change AI providers, we’ll update this page and notify users in advance of the change.

10. Future AI features

AI is core to YachtSync. Things we’re considering or actively building:

• Bulk scanning. Drop in 50 PDFs at once and have AI process them in the background.
• Premium scanning. Higher-accuracy or larger-context AI models for paid tiers.
• Natural language search. Ask “what expires in the next 60 days?” rather than tapping through filters.
• Career assistant. Suggest training pathways based on your certificates and goals.

Any new AI feature will follow the same disclosure rules: we’ll update this page first, explain what data is processed, and give you the option to opt out where reasonable.

11. Contact

For questions about how YachtSync uses AI:

Email: legal@yachtsync.uk
Post:
YachtSync Ltd
66 Paul Street
London EC2A 4NA
United Kingdom