AI Governance Policy

Last updated: June 2025

1. Purpose

This AI Governance Policy establishes the principles, processes, and controls by which MARQX WISDOM (OPC) PRIVATE LIMITED ("MarqX") develops, deploys, and monitors AI systems — both internally and within the AI Systems it delivers to enterprise customers.

2. Scope

This Policy applies to all AI systems developed, operated, or procured by MarqX, including machine learning models, large language models, AI agents, and automated decision systems that have a material impact on individuals, enterprises, or downstream systems.

3. Principles

MarqX is committed to the following AI principles:


Fairness: AI systems shall not discriminate against individuals or groups on the basis of protected characteristics. Fairness testing is mandatory before deployment.


Transparency: The purpose, capabilities, and limitations of every AI system shall be documented and available to affected stakeholders.


Accountability: Every AI system has a named owner responsible for its performance, governance, and compliance.


Privacy by Design: AI systems are designed to minimise personal data use and apply privacy-preserving techniques where applicable.


Human Oversight: High-stakes AI decisions require human review before action is taken.


Non-maleficence: We will not develop or deploy AI systems whose primary purpose is to harm, deceive, or discriminate.

4. Risk Classification

AI systems are classified into four risk tiers:


Critical: AI systems making or directly influencing decisions with irreversible consequences for individuals (e.g., credit decisions, medical triage). Require full governance sign-off before deployment.


High: AI systems with significant business impact or material effect on individuals. Require fairness testing, explainability documentation, and human-in-the-loop controls.


Medium: AI systems with moderate impact. Require standard evaluation, monitoring, and documentation.


Low: AI systems with minimal risk (e.g., content summarisation, internal search). Require basic logging and periodic review.

5. Model Governance

Every AI model deployed through the MarqX platform is subject to:


  • **Model registration**: Entry in the MarqX model inventory with purpose, data provenance, version, owner, and risk classification
  • **Pre-deployment evaluation**: Performance, fairness, and bias testing on held-out datasets
  • **Evidence pack**: Documentation of training data, evaluation results, known limitations, and governance sign-off
  • **Post-deployment monitoring**: Continuous performance and drift monitoring with automated alerts
  • **Periodic review**: Scheduled re-evaluation at intervals appropriate to the risk classification

    • 6. Human-in-the-Loop

    For Critical and High risk AI systems, human-in-the-loop controls are mandatory. These include:


  • Configurable approval gates before AI-recommended actions are executed
  • Override capabilities available to authorised personnel at all times
  • Full audit trail of human interventions, approvals, and overrides

    • 7. Incident Response

    AI incidents (including unexpected outputs, bias detection, performance degradation, and security events) are handled under our incident response framework:


  • P1 AI incidents: Response within 1 hour, model suspension if necessary
  • Root cause analysis within 72 hours
  • Remediation plan and governance review before reinstatement
  • Affected parties notified within statutory timelines

    • 8. ISO 42001 Alignment

    This Policy and our AI Management System are designed to align with ISO/IEC 42001:2023, the international standard for AI management systems. We conduct annual gap assessments and maintain evidence packs suitable for external audit.

    9. Review and Updates

    This Policy is reviewed annually and following any material change to our AI development practices, applicable laws, or industry standards. Updates are communicated to all relevant teams and enterprise customers.