Legal Center

AI Features and Image Integrity Policy

Status: Draft v0.9 — 11 August 2026 — legal and operational review required before publication. Current provider: Heritage Timepieces AB, org. no. 559491-1157, VAT SE559491115701, Kungsgatan 2C, 223 50 Lund, Sweden. Entity notice: Klocktech AB is under registration and is not the current contracting party. Registration alone does not transfer an agreement. Contacts: legal@heritagetp.com (legal/privacy) · info@heritagetp.com (general).

1. Scope

This policy applies to AI-assisted image preparation, description drafting, technical-attribute suggestions, translation, pricing suggestions, aggregate market insights and any conversational assistance offered in Klocktech. It supplements the SaaS Terms, AUP and DPA.

2. Intended use

AI features are assistive tools. They may:

  • isolate a dealer-photographed watch from its background;
  • crop, resize and format an image;
  • make restrained, non-generative tonal corrections under the controls in section 4;
  • apply a preset or dealer-uploaded background without changing the watch;
  • create destination-specific watermarked or unwatermarked variants;
  • draft descriptions, short history text and translations from verified fields;
  • suggest structured attributes from Customer-provided information or Permitted Reference Data; and
  • provide price calculations or non-identifiable, non-singlable aggregate market indicators from approved sources.

They are not intended to authenticate, appraise, certify, inspect or determine legal ownership of a watch.

3. Non-negotiable source boundary

AI features must not browse, scrape, retrieve, ingest or reproduce another dealer's inventory, listings, photos, text, serials, prices, seller identity or availability. AI enrichment must use Customer Data and Permitted Reference Data only. A licence, consent or technically accessible interface does not make another dealer's item-level data permissible. An AI or model provider must not be given a browsing, generic URL-fetch or marketplace-search tool in the Klocktech generation path.

The Customer must not prompt or instruct an AI feature to bypass this rule. Market insight may return aggregate statistics, source category, timestamp, sample size and confidence, but not identifiable or singlable comparable listings. Risk-based cohort thresholds, rare-combination suppression, overlapping-query controls and no-underlying-row rules must prevent reconstruction of seller- or watch-level data.

4. Photographic truthfulness

Klocktech must preserve an immutable original of each dealer-uploaded image and record the processing version, segmentation mask, operations and approver. Generative processing is confined to pixels outside the watch segmentation mask. Narrowly constrained deterministic tonal correction may touch watch pixels only where it cannot invent, erase, repair, beautify or obscure condition evidence. Cropping must not omit relevant damage or included components. In particular, the system and user must not generate, remove, replace, sharpen into existence, hide or beautify:

  • dial printing, hands, indices, bezel markings or crystal;
  • case, crown, pushers, bracelet, strap, clasp or movement;
  • engravings, hallmarks or reference information;
  • scratches, dents, oxidation, patina, polishing evidence or damage; or
  • accessories, box, papers, tags or components not present in the source image.

The sole narrow exception is a dedicated serial-privacy derivative. It may blur or opaque-mask the serial region without modifying the immutable original, synthesising replacement characters or concealing other condition evidence. The Customer chooses backend-only or a public mask showing no more than three original characters. Full serials must not be included in public image metadata, filenames, text or public channel fields.

Every processed image must pass an automated comparison appropriate to the operation and be previewed side-by-side against the original before publication. If segmentation or comparison confidence is insufficient, Klocktech must fall back to the original or require manual processing outside the Service. The Customer is responsible for approval and accurate condition disclosure. Klocktech may block or withdraw manipulative output.

5. Text and translation safeguards

Generated text is a draft. It must not state authenticity, provenance, service history, rarity, ownership, warranty, condition, full-set status or included items unless supported by verified Customer fields. Historical context and translations may be incomplete or wrong. A knowledgeable user must review every material statement and destination language.

6. Prices and market information

AI suggestions, currency conversions and aggregate market indicators are not valuations, investment recommendations or promises of sale. The Customer selects the market, reviews source and timestamp information, and approves the final price. Identifiable competitor observations are prohibited.

7. Inputs, outputs and model training

The Customer confirms it has rights and lawful authority for prompts, photos, logos, backgrounds, product facts and personal data. The Customer retains its input rights and may use output as provided in the SaaS Terms, subject to law and third-party rights.

Klocktech will not use identifiable Customer inventory, images, prompts, serials, prices, listings or output to train a general-purpose or cross-customer AI model or create cross-dealer benchmarking, including where a Customer offers to opt in. Only irreversibly anonymised and sufficiently aggregated operational statistics that cannot identify or single out a Customer, person, watch, listing or confidential commercial pattern may be used across customers. Any external AI vendor that processes personal data on the Customer's behalf must be listed as a subprocessor and bound by the DPA, and must be contractually prohibited and technically configured from training on Customer Data. Actual vendor, location, retention and training settings must be confirmed in the Subprocessor Register before production use.

8. Human control and accountability

AI must not make the final decision to publish, price, represent condition, reserve, sell or delist a watch. Users must be able to edit, reject and revert output. Material AI operations, source fields, model or tool version and approval should be recorded in the audit trail.

9. AI transparency

Klocktech will inform users when they are interacting with an AI feature where that is not obvious. It will implement machine-readable marking or disclosure for generated or manipulated content where legally required and technically applicable. The Customer must preserve required markers and make any destination-facing disclosure required for its use.

The treatment of background replacement and standard image editing under Article 50 of the EU AI Act depends on the exact technical operation and whether the input or its meaning is substantially altered. Klocktech will not claim an exemption until that feature has been assessed. Where uncertain, the safer default is transparent disclosure that an image was AI-assisted while confirming that the watch itself was not altered.

10. AI literacy and change control

Personnel operating or supervising AI features must receive role-appropriate guidance on capabilities, limits, image integrity, privacy, source restrictions and human review. Klocktech will reassess legal classification, documentation and transparency when a feature, model, intended purpose or risk materially changes.