1. Purpose
This policy sets out Murzo Group's approach to AI training data, dataset sourcing, model input, model output, prompts, embeddings, synthetic data, fine-tuning, evaluation data, and generated content.
The purpose is to reduce legal, privacy, copyright, confidentiality, bias, security, product, and reputational risk when using or developing AI systems.
2. Scope
This policy applies to datasets, prompts, uploads, documents, images, audio, video, source code, product data, customer data, supplier data, worker data, farm data, food data, cultural property data, security data, public web data, licensed data, synthetic data, and AI outputs used by or for Murzo Group.
It applies to internal tools, third-party AI tools, hosted models, foundation models, chatbots, automation, analytics, computer vision, voice tools, robotics, recommendation tools, and model evaluation.
3. Approved Data Sources
Data used for AI must have a suitable business purpose and lawful basis or permission where required. Murzo Group should prefer data that is owned by Murzo Group, licensed for the intended use, lawfully obtained, public in a way compatible with the intended use, properly anonymised, or provided under written terms that allow the relevant AI activity.
Data must not be assumed safe to use merely because it is online, publicly visible, scraped, generated by AI, shared by a contractor, or technically accessible.
4. Prohibited and Restricted Data
The following must not be used for AI training, fine-tuning, prompting, testing, embeddings, or model enrichment unless specifically approved and lawful:
- Personal data, special category data, children's data, biometric data, worker data, CCTV, body-worn camera footage, recruitment data, customer complaints, or health information
- Client confidential information, trade secrets, passwords, API keys, contracts, legal advice, financial data, security data, vulnerability information, source code, or unpublished business plans
- Copyright material, design files, fashion artwork, photographs, audio, video, datasets, cultural property records, supplier catalogues, or scraped content where rights are unclear
- Food safety, allergen, traceability, animal, farming, biosecurity, product safety, or regulator information where misuse could cause harm
5. Personal Data and Privacy
AI activity involving personal data must follow Murzo Group's Data Protection Policy and applicable UK, EU, Chinese, and other local data protection laws. Privacy risk, transparency, data minimisation, accuracy, fairness, security, retention, individual rights, automated decision-making, and cross-border transfer rules must be considered where relevant.
Personal data should not be entered into public or consumer AI tools unless approved for that tool, purpose, data type, and jurisdiction.
6. Copyright, IP and Licensing
AI training, fine-tuning, retrieval, generation, and output use may create copyright, database right, trade mark, design right, confidentiality, moral rights, passing off, and contract risk.
Murzo Group must not claim ownership, originality, exclusivity, authorship, or freedom to use AI outputs unless the underlying tool terms, inputs, human contribution, and legal position support that claim. Outputs intended for commercial use, branding, product design, fashion, marketing, cultural property, publication, customer advice, code, or legal documents may require review.
7. Output Reliability and Human Review
AI outputs may be inaccurate, incomplete, outdated, biased, fabricated, infringing, unsafe, or unsuitable for the context. Outputs must not be relied on as the sole basis for legal, safety, medical, financial, employment, product, food safety, security, cultural property, export control, or customer-impacting decisions.
Human review is required where an AI output may affect rights, safety, compliance, customer commitments, product quality, public communications, contracts, money, employment, regulated activity, or reputation.
8. Model Improvement and Feedback Loops
Where Murzo Group provides feedback, corrections, user inputs, evaluation data, or operational data to an AI supplier, the supplier's rights to use that information must be understood. Sensitive data must not be used to improve third-party models without suitable approval.
Feedback loops must be monitored where they could reinforce bias, error, unsafe recommendations, discriminatory outcomes, hallucinations, manipulation, or poor quality.
9. Third-Party Liability and Unauthorised Use
Third parties, contractors, suppliers, creators, agencies, affiliates, marketplaces, customers, and platform operators must not use Murzo Group data to train, fine-tune, enrich, test, or prompt AI systems unless authorised in writing.
To the fullest extent permitted by law, Murzo Group does not accept responsibility for unlawful data sourcing, copyright infringement, privacy breach, confidentiality breach, model contamination, biased output, or unsafe use caused by third parties acting outside written approval.
10. Review and Evidence
Murzo Group may keep proportionate evidence of data source decisions, licences, approvals, supplier terms, output reviews, and escalations where needed for legal, privacy, IP, security, customer, or dispute purposes.
This policy should be reviewed when AI tools, law, supplier terms, data sources, model use, product lines, or international operations change.
