AI Transparency & Fair-Use Policy

Effective date: 31 October 2026

English is the source and governing version of this AI Policy. If any translation differs from the English version, the English version prevails to the fullest extent permitted by applicable law.

1. Who we are and scope

This AI Policy is issued by AMYA AFRICA LTD, a company registered in England and Wales under company number 17469906, with its registered office at 128 City Road, London, EC1V 2NX, United Kingdom ("Amya", "we", "us" or "our").

Amya is a cross-border business-to-business (B2B) marketplace connecting African commercial merchants directly with manufacturers, exporters and suppliers worldwide, particularly in Asia. Because buyers and suppliers typically operate in different countries and languages, Amya integrates artificial intelligence ("AI") to assist with translating, transcribing, drafting, summarizing, moderating and searching marketplace content.

This Policy explains each AI feature across the Amya websites and the Amya Android application: what it does, the limitations of AI outputs, where human review and confirmation apply, how data is handled by our enterprise AI processors, and how AI outputs are labelled. It supplements the Amya Terms of Use (in particular Section 8, AI-assisted features and transparency), the Acceptable Use Policy and the Privacy Notice.

Contact:

Questions about AI features: support@amya.africa
Privacy enquiries: privacy@amya.africa

2. Core AI principles

Clear disclosure: You are informed when you are interacting with an AI assistant, and AI-generated or machine-translated outputs are labelled clearly in the interface (Section 6).
AI suggests; humans decide: AI never sends a message, submits an RFQ, issues a quotation, publishes a product listing or modifies your live marketplace data autonomously. Every external action proposed by an AI assistant requires your explicit confirmation (Section 4.1). Only human Amya staff can refuse organization verification, and every moderation appeal is decided by a human reviewer.
Preservation of originals: Machine translation never overwrites the original source message or listing text, and the AI photo studio never deletes your original uploaded product photograph.
Mandatory human verification: AI outputs can be inaccurate or incomplete; your organization must review and verify all critical trade figures and terms before relying on them (Section 3).
No training of public models on your content: Your private business data and documents are not used to train public AI foundation models (Section 5).
Safe fallback: If an AI feature is unavailable or fails, Amya displays the original unmodified content rather than guessing or blocking access to the original text.

3. AI limitations and your responsibility to verify

AI translations, transcripts, summaries, drafts, image edits, semantic search matches and assistant responses are generated probabilistically by machine-learning models, not human trade advisers. They can contain errors, omissions or hallucinations, and may mistranslate technical specifications, quantities, units of measurement, currencies, Incoterms® or delivery schedules.

Before relying on any AI output or entering into any commercial commitment, your organization must independently verify the output and confirm all binding quantities, specifications, prices, currencies, payment instructions and delivery terms directly with your counterparty. AI outputs never constitute legal, tax, customs, regulatory or financial advice, and an AI conversation summary or machine translation does not alter or form part of your sale contract. Your organization remains solely responsible for every listing, RFQ, quotation and message it chooses to publish or send, including content prepared with AI assistance.

4. Our AI features

Access to specific AI features and usage quotas depends on your organization's role and subscription plan, as detailed on the Amya plans page.

4.1 AI assistant and supplier copilot (text and voice)

Capabilities: Helps users navigate and work on Amya. For buyers, it can search and compare products, identify matching suppliers and prepare draft RFQs. For suppliers on eligible plans, it can summarize active conversations and RFQs, draft replies or quotations, and surface catalogue performance and unmet-demand insights.
Strict permission boundaries: The assistant can access only records that your signed-in user account is already authorized to view under Amya's role-based access controls.
Mandatory human confirmation: The assistant cannot transmit messages, RFQs or quotations on its own. Whenever it prepares an action that would reach another user or alter marketplace data, it presents a draft proposal for your review. Nothing is executed unless you explicitly click Confirm (you may edit or Reject the proposal first, and unconfirmed proposals expire automatically after 24 hours).
Voice interaction: Where you speak to the assistant, voice audio is streamed to our AI processor for real-time speech processing; Amya does not store session voice audio, retaining only the resulting text transcript within your assistant thread alongside usage duration metrics.
Retention: Assistant conversation threads are retained until you delete the thread or delete your account.

4.2 Machine translation of chat messages

Capabilities: Where a conversation includes machine translation, messages are translated automatically into each participant's preferred language.
Original text preserved: The sender's original wording is always retained alongside the translation, allowing participants to view both versions (and toggle automatic translation off where supported in the interface).
Retention: Translations are stored with the message record and deleted when the conversation is deleted.

4.3 Translation of listings, RFQs, quotations and profiles

Capabilities: Product listings, RFQs, quotations and supplier profiles are machine-translated into Amya's supported languages so international buyers and suppliers can browse and source in their selected language.
Human translations respected: Machine-translated fields are recorded and labelled as machine-generated, and never overwrite a translation manually authored by a user.

4.4 Uploaded voice notes: transcription and "listen in my language"

Capabilities: When a user records and sends a voice note in a chat conversation, AI transcribes the audio, detects the spoken language (skipping translation if language-detection confidence is too low), translates the transcript, and allows recipients to play back a synthesized speech reading of the translation ("listen in my language").
Retention: Uploaded voice notes, their transcripts, translations and synthesized audio files are stored as part of the chat conversation for participant playback and deleted when the conversation or account is deleted.

4.5 Live interpreted marketplace calls (currently disabled)

Live interpreted calls between buyers and suppliers are not currently enabled. If and when launched:

an AI interpreter participant ("Interpreter") will provide real-time speech translation during the call;
every call participant will be notified and required to consent to AI interpretation before joining;
Amya will never record or store call audio; and
the AI interpreter's text captions and translations will be saved as call transcript segments only if every call participant explicitly consents to keeping them (solely for commercial dispute handling, and deleted when the conversation or account is deleted).

4.6 Listing drafts from product photos or voice dictation

Capabilities: Suppliers can upload product photographs or dictate a spoken description for AI to generate a draft product listing (suggested title, description, attributes, image alt-text and candidate HS tariff code). Uncertain fields are left blank.
Human review required: Drafts are never published automatically. The supplier must review, edit and confirm every field before submitting the listing. Any suggested HS code is an indicative aid only; the supplier and buyer remain solely responsible for lawful customs tariff classification.

4.7 AI photo studio

Capabilities: Suppliers can use predefined studio tools to clean product photo backgrounds, adjust lighting, upscale resolution ("AI-enhanced"), or place the product into a studio background scene ("AI-generated"). The studio operates solely on fixed commercial enhancement rules and does not generate arbitrary free-text imagery.
Safeguards: Your original product photo is always preserved, and an AI-generated scene cannot be set as the primary cover image of a product listing. All studio outputs pass automated safety moderation before display, carry visible AI labels where displayed, and preserve machine-readable provenance metadata (such as Google SynthID watermarks and C2PA content credentials where supported by the model).
Accuracy obligation: Enhanced or scene-placed images must remain an accurate, non-misleading representation of the actual physical goods you supply.

4.8 Profile descriptions, voice RFQs, suggested replies and summaries

Drafting tools: AI can draft a supplier "About us" profile description, convert a buyer's spoken dictation into a draft RFQ for review before saving, propose short chat reply options for eligible suppliers (stripping personal names/contact details before prompt submission and discarding any suggestion containing payment details or external links), and generate non-binding conversation summaries. Nothing is saved or sent without user action.

4.9 Semantic search and visual search

Semantic matching: Product listings, RFQs and search queries are converted into numerical vector embeddings so search and RFQ matching can identify relevant commercial results by meaning as well as keywords. Suppliers flagged as "Matched by Amya" are automated relevance suggestions, never commercial endorsements.
Search by photo: Where a buyer searches using an image, the image is screened by safety moderation and analysed by our AI processor to generate a descriptive text search query; the uploaded search photo itself is not stored permanently by Amya.

4.10 Automated content moderation

AI models screen platform text (listings, profiles, RFQs, quotations and review replies) and images (catalogue, profile, studio and chat media) against violations of the Acceptable Use Policy. Private chat message text and voice-note transcripts are not screened automatically; they are reviewed by our staff when reported. Content rejected by automated checks is withheld or removed with a formal statement of reasons disclosing the automated decision and providing a right of appeal to a human reviewer. For moderated text, Amya logs a cryptographic hash of the text and the moderation result rather than duplicating the message body in moderation logs.

4.11 Organization verification pre-check

Where enabled, an AI model inspects submitted business-registration certificates and representative identity documents to compare names and validity dates against the organization profile and any available corporate registry, assigning a provisional status ("checked automatically, review pending") when no issue is detected. AI never refuses verification: all final approvals and refusals are decided by human Amya staff, no facial recognition or biometric processing is used, and the identity-document image is permanently deleted promptly after the staff decision.

4.12 User feedback on AI outputs

Users may rate AI outputs (helpful / not helpful) and submit optional comments to help us monitor and improve feature quality. AI feedback records are purged 365 days after creation.

5. How our AI service providers handle your data

Enterprise AI processors: Amya uses Google Gemini on Vertex AI (Google Cloud Platform) for language, translation, transcription, assistant, vision, studio, embedding, moderation and document pre-check functions, and Google Cloud Text-to-Speech for speech synthesis. Google acts as Amya's data processor under binding Google Cloud data-processing terms.
Data minimization: Only the inputs necessary for the specific feature invoked are transmitted to the AI processor. Do not submit unnecessary third-party personal data or sensitive personal information in AI prompts or uploads.
No training of public models: Under our Google Cloud data-processing terms, our AI processors are contractually prohibited from using Amya customer content to train or fine-tune their public foundation models, and Amya itself does not train or fine-tune foundation models on your content.
Payload retention: For each background AI task, Amya logs the feature type, model identifier, prompt version and token/minute usage for plan-allowance metering. Raw task input and output payloads are cleared 90 days after task completion, while outputs incorporated into your workspace (such as saved listings, chat translations or assistant threads) follow the retention rules in the Privacy Notice.
Processing locations and transfers: While Amya's primary database is hosted on Google Cloud in the Netherlands (europe-west4), AI model inference may be processed on Google Cloud endpoints in the European Economic Area, the United Kingdom, the United States or global Vertex AI endpoints depending on model availability. All international data transfers are protected by the standard contractual safeguards described in Section 6 of the Privacy Notice.
Service limits: AI features are subject to your plan's usage allowances and fair-use rate limits, and may be throttled, updated or suspended by Amya for safety, quality or operational reasons.

6. Transparency, labelling and EU AI Act alignment

In accordance with UK data-protection principles and Article 50 of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) where applicable:

Human-facing AI interactions: AI assistants and any future live call interpreter are clearly identified as AI systems so users always know they are interacting with machine intelligence rather than a human.
Labelling of synthetic and translated outputs: Machine translations, synthesized speech audio, AI conversation summaries and AI-enhanced or AI-generated studio images are labelled as AI-generated in the interfaces where they are presented, and machine-readable provenance metadata is preserved on studio images where provided by the underlying model.
Disclosure of automated moderation: Moderation statements of reasons explicitly state when a restriction was triggered or taken by automated tools, and every affected user has the right to obtain human review by appealing the decision.

You must never strip, obscure or falsify any AI label, watermark or provenance metadata, or misrepresent AI-generated content as human-verified certification. Amya never uses AI for biometric categorization, emotion recognition or social scoring.

7. Acceptable use of AI features

When using Amya's AI tools, you must not:

attempt to bypass safety filters or manipulate model behaviour through adversarial prompts ("prompt injection" or jailbreaks) embedded in messages, listings or uploaded documents;
use AI translations, drafts or studio images to deceive a counterparty about the nature, origin, certification or commercial terms of your goods;
use AI tools to probe unauthorized data or scrape platform content; or
upload third-party personal data or confidential material that you have no lawful right to process.

8. Changes to this AI Policy and contact

We will update this AI Policy when we introduce or modify AI features, change AI processors, or reflect new legal standards, publishing the updated version with a new effective date in accordance with the Terms of Use.

AMYA AFRICA LTD (Company No. 17469906) Registered office: 128 City Road, London, EC1V 2NX, United Kingdom

AI features and general support: support@amya.africa
Privacy and personal-data rights: privacy@amya.africa