Responsible AI Policy
Published at: www.metadat-ai.com/legal/policies/responsible-ai
Document reference: MAI-RAI-001 Version: v1.3 Last updated: October 2026 © METADAT-AI LTD (Company No. 17486598)
| This policy is a Living Document: the version published at the URL above is current, and updates take effect under the Living Document rules of any agreement that incorporates it. |
1. Purpose and Scope
METADAT-AI LTD (“Metadat-AI”) develops and operates AI systems that analyse video, audio and images. This policy sets out how Metadat-AI aims to develop, deploy and operate those systems responsibly, ethically and lawfully. It applies to Metadat-AI’s directors, employees and contractors involved in developing, testing, deploying, operating or selling the Platform, and to the selection and use of third-party AI models and components. The measures in it are proportionate to Metadat-AI’s size and risk profile and will develop as the business grows.
2. Principles
(a) Lawfulness: AI systems are developed and operated in line with applicable law, including the UK GDPR, the EU AI Act where it applies, and sector-specific rules.
(b) Transparency: Metadat-AI tells customers that the Platform uses AI and machine learning, explains the general nature and purpose of those systems and their material limitations (including accuracy limitations and the potential for error), and publishes the licensing basis of open-source components in the Open Source Licensing Statement. It does not make false or misleading claims about the capabilities, accuracy or reliability of its AI systems in marketing, sales or contractual materials.
(c) Fairness: Metadat-AI takes reasonable steps to identify and reduce bias in its AI systems (section 4).
(d) Human oversight: the Platform is designed to support human judgement, not replace it (section 5).
(e) Accountability: Metadat-AI keeps clear internal accountability for AI development decisions (section 9).
(f) Privacy and security: AI systems respect individuals’ privacy and are protected by appropriate security controls (section 6).
(g) Proportionality: the governance applied to an AI system is proportionate to the risk it poses.
3. Model Selection, Training Data and Test Content
Open-source models. Metadat-AI aims to use only open-source models and frameworks whose licences permit commercial use, and does not use a model or model weights licensed only for non-commercial use in a commercial service unless it first obtains a commercial licence. It aims to keep an internal register of the models and frameworks used in the Platform and their licences, and to review it when a new component is adopted. Where a component is subject to copyleft conditions (such as the GNU General Public Licence, “GPL”), Metadat-AI aims to meet those conditions before commercial deployment. Further information is in the Open Source Licensing Statement at www.metadat-ai.com/legal/policies/oss.
Training data. Where Metadat-AI trains, fine-tunes or adapts AI models, it aims to use only datasets released under licences that permit machine learning training, content it owns or has created, and content it has licensed for that purpose, and to keep records of the datasets used and their licence basis, reviewed annually.
Customer data. Metadat-AI does not use customer or end-user data to train, fine-tune or improve its general AI models. A Custom Model is trained only on that customer’s own content, on its instructions, and is used only for that customer.
Demonstration and test content. Development and testing use synthetic, licensed or Metadat-AI-owned content, not real customer data. Demonstrations and trials follow the Demonstration and Trial Policy: synthetic data, or real data only with the permissions required under the Data Processing Agreement and, for biometric data, the Biometric and Facial Recognition Processing Addendum.
4. Fairness and Bias
AI systems can produce biased or discriminatory outputs, particularly where training data is unrepresentative. Metadat-AI aims to:
(a) consider the diversity and representativeness of training data during model development;
(b) test model outputs for material performance differences across demographic groups where relevant to the use case;
(c) document known limitations and biases of models deployed on the Platform; and
(d) tell customers about material limitations of AI outputs and, where outputs may materially affect individuals, the intended use, known limitations and recommended human oversight measures.
5. Human Oversight
AI is never fully accurate. AI outputs are probabilistic, cannot be relied on as the sole basis for any decision, and must be reviewed by a human before any decision is made on them. Metadat-AI keeps human oversight of its own AI development and deployment decisions, does not deploy AI where fully automated decision-making would have legal or similarly significant effects on individuals without appropriate safeguards, and designs the Platform so that customers can review, correct and override AI outputs.
6. Privacy and Security
AI development and operation follow data protection law and Metadat-AI’s internal policies. Personal data is not used in model training without a lawful basis; privacy by design and data minimisation are applied; and AI systems that process personal data are subject to a data protection impact assessment where the UK GDPR requires one.
AI systems are covered by Metadat-AI’s internal security policies. Metadat-AI aims to protect them against adversarial attacks and model inversion attacks, to store trained model weights and related artefacts securely with restricted access, and to monitor production outputs for anomalous or unexpected behaviour.
7. EU AI Act
Metadat-AI places its AI system on the market in its own name and monitors the phased implementation of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744): the prohibitions apply from 2 February 2025, the transparency obligations in Article 50 from 2 August 2026, and the high-risk obligations relevant to the facial recognition and age and gender estimation features from 2 December 2027. Where a feature is or becomes subject to specific requirements of that Act (for example those for high-risk AI systems), Metadat-AI takes reasonable steps to meet the requirements that apply to it, including conformity assessment, technical documentation and transparency obligations. Metadat-AI gives deployers Instructions for Use for these features.
8. Customer Use and Incidents
Customer use of the Platform is governed by the Acceptable Use Policy (www.metadat-ai.com/legal/policies/aup). AI-related incidents (including material AI errors, harmful outputs, bias complaints and misuse) are reported to the CEO and handled under Metadat-AI’s internal policies; affected customers are notified where required.
9. Governance and Review
The CEO has overall accountability for responsible AI at Metadat-AI. Decisions to deploy new AI systems or materially change existing ones are reviewed against this policy before they are made. Metadat-AI aims to review this policy annually, and earlier if AI regulation, data protection law or guidance, or Metadat-AI’s technology changes materially.
10. Contact
Questions about this policy should be sent to legal@metadat-ai.com.
MAI-RAI-001 v1.3. © METADAT-AI LTD (Company No. 17486598). This document forms part of the Metadat-AI legal framework published at www.metadat-ai.com/legal/.