How to build an AI governance framework compliant with the EU AI Act
Fashion brands must now assign clear roles, maintain risk registers, and create audit trails under the Digital Omnibus amendments to the EU AI Act.

KEY TAKEAWAYS Summary by the editors
- The Digital Omnibus amendments to the EU AI Act introduce stricter requirements for AI governance frameworks in fashion businesses operating in the EU.
- Brands must designate specific roles including an AI governance lead and establish cross-functional oversight committees.
- Risk registers must be maintained and updated quarterly to document AI system classifications and mitigation measures.
- Audit trails documenting all AI system decisions, training data, and model changes are now mandatory for high-risk applications.
Fashion brands deploying AI systems in the European Union must establish formal governance frameworks that assign clear responsibilities, maintain risk registers, conduct quarterly reviews, and preserve audit trails. These requirements stem from the Digital Omnibus amendments to the EU AI Act, which came into force in stages throughout 2024 and 2025, with full compliance deadlines extending into 2026 and 2027. For fashion businesses using AI in demand forecasting, pricing, customer profiling, or supply chain management, the governance structure determines whether systems meet regulatory standards.
What does the Digital Omnibus require from fashion brands?
The Digital Omnibus package, adopted by the European Parliament in 2024, amended multiple EU digital regulations including the AI Act to harmonise enforcement and reduce compliance fragmentation. The amendments clarify obligations for AI deployers,the businesses that use AI systems, rather than just the developers who build them. Fashion retailers and brands that deploy AI tools now face explicit requirements to document system purposes, assess risks, and demonstrate human oversight.
The AI Act classifies systems by risk level. High-risk applications include AI used in employment decisions, creditworthiness assessments, or systems that significantly influence consumer behaviour through personalised pricing or targeted offers. Many customer-facing AI tools in fashion,recommendation engines, dynamic pricing algorithms, and automated credit checks,fall into categories requiring stringent governance. Limited-risk systems, such as basic chatbots, require transparency disclosures but lighter governance. Minimal-risk applications, like simple inventory trackers without predictive functions, face few obligations.
Who should lead AI governance in a fashion business?
Effective governance starts with role assignment. Brands should designate an AI governance lead at senior management level, typically reporting to the chief technology officer, chief information officer, or chief risk officer. This individual coordinates compliance activities, liaises with regulators, and chairs the AI governance committee. The governance committee should include representatives from legal, IT, data privacy, risk management, merchandising, and customer experience teams to ensure cross-functional oversight.
Each AI system deployed should have a named system owner responsible for documenting its purpose, monitoring performance, and escalating issues. For example, if a fashion retailer uses an AI tool for demand forecasting, the head of planning or merchandising might serve as system owner. If the brand deploys AI for recruitment screening, the human resources director typically owns that system. This distributed ownership model ensures that business functions using AI remain accountable for its outcomes.
How should brands structure their AI risk register?
A risk register is a living document that catalogues every AI system in use, its risk classification, intended purpose, data sources, and mitigation measures. The register should be updated quarterly or whenever a new system is deployed or an existing system is modified. Each entry must include the system name, vendor or developer, deployment date, risk category, and responsible owner.
The register should document specific risks identified for each system. For a dynamic pricing algorithm, risks might include discriminatory outcomes, opacity in price-setting logic, or potential consumer manipulation. Mitigation measures could include regular bias audits, human review of price changes above certain thresholds, and transparent communication of pricing policies. For AI-powered hiring tools, risks include unfair filtering of candidates, while mitigations might involve regular fairness testing and human oversight of shortlisting decisions.
| System | Risk level | Owner | Key risks | Mitigations |
|---|---|---|---|---|
| Demand forecasting | Limited | Head of Planning | Data quality, overfitting | Monthly accuracy reviews, data validation |
| Dynamic pricing | High | E-commerce Director | Discrimination, transparency | Bias audits, price cap rules, disclosure |
| Recruitment screening | High | HR Director | Unfair filtering, bias | Fairness testing, human review |
| Product chatbot | Limited | Customer Experience Lead | Misinformation, privacy | Human escalation, data controls |
What must quarterly reviews cover?
The AI governance committee should convene at least quarterly to review the risk register, assess system performance, evaluate incidents, and update policies. Each review should examine whether systems remain fit for purpose, whether risk classifications have changed, and whether mitigations are working. Minutes of these meetings form part of the audit trail and may be requested by regulators during inspections.
Quarterly reviews should include performance metrics for each high-risk system. For instance, if a fashion brand uses AI to recommend products, the review might assess conversion rates, customer complaints, and any evidence of biased recommendations across demographic groups. If a system's error rate rises or complaints increase, the committee may decide to suspend the system, retrain the model, or implement additional safeguards. The review should also confirm that system owners have completed required documentation and training.
How do you create compliant audit trails?
Audit trails document the lifecycle of each AI system from procurement or development through deployment, monitoring, and decommissioning. For high-risk systems, brands must retain records of training data, model architecture, validation testing, deployment approvals, performance logs, and any modifications. These records must be stored securely and made available to regulators on request.
Documentation should include data provenance,where training data originated, how it was labelled, and whether it contains personal information. If a fashion retailer trains a demand forecasting model on historical sales data, the audit trail should record the data sources, any preprocessing steps, and how the model was validated before deployment. If the brand later updates the model with new data, that change must be logged with dates, reasons, and approvals.
For customer-facing systems, audit trails should capture individual decisions where feasible. If an AI system denies credit or rejects a job applicant, the brand must be able to explain the decision, identify which data inputs influenced the outcome, and demonstrate that the system operated as intended. This level of traceability requires logging not just aggregate performance but decision-level data, which may necessitate technical changes to AI platforms.
What happens if a brand fails to comply?
Non-compliance with the EU AI Act can result in administrative fines. The regulation sets maximum penalties based on the type of infringement and the company's global annual turnover. Serious violations, such as deploying a prohibited AI system, can attract fines of up to 7 per cent of global turnover. Failures in governance, documentation, or transparency can result in fines of up to 3 per cent of turnover or set amounts, whichever is higher.
Beyond fines, reputational damage and operational disruption present significant risks. Regulators may order a brand to suspend or withdraw non-compliant AI systems, disrupting merchandising, pricing, or customer service operations. For fashion businesses with thin margins, even a temporary halt to AI-driven demand forecasting or pricing tools could affect profitability. Establishing robust governance frameworks now reduces these risks and positions brands to adopt AI responsibly as the technology evolves.
- Designate an AI governance lead and convene a cross-functional committee
- Catalogue all AI systems in a risk register with quarterly updates
- Assign a system owner to each AI application with clear accountability
- Document training data, model changes, and decisions for high-risk systems
- Conduct quarterly governance reviews with minutes and action logs
- Ensure audit trails are secure, complete, and accessible for regulatory requests
Frequently asked questions
Which fashion AI systems are considered high-risk under the EU AI Act?
High-risk systems include AI used in employment decisions, creditworthiness assessments, and tools that significantly influence consumer behaviour through personalised pricing or targeted offers. Many recommendation engines, dynamic pricing algorithms, and automated hiring tools fall into this category and require stringent governance, risk registers, and audit trails.
How often must fashion brands update their AI risk registers?
Risk registers should be updated quarterly or whenever a new AI system is deployed or an existing system is modified. Each update should document changes in system purpose, risk classification, data sources, or mitigation measures, and these updates form part of the mandatory audit trail.
What penalties can fashion brands face for non-compliance with the EU AI Act?
Non-compliance can result in administrative fines of up to 7 per cent of global annual turnover for serious violations or up to 3 per cent for governance failures. Regulators may also order brands to suspend or withdraw non-compliant systems, causing operational disruption and reputational damage.
