AI Auditing
AI Auditing is the systematic examination of AI systems to assess accuracy, fairness, transparency, and compliance with ethical and regulatory standards, evaluated within fashion and sustainability contexts including algorithmic sourcing, demand forecasting, automated labour monitoring, and supplier-scoring tools.
The concept of auditing algorithmic systems predates the term AI auditing itself. Its intellectual roots lie in software verification practices of the 1970s and 1980s, when computer scientists first grappled with how to validate that automated systems performed as specified. These early frameworks were technical, not ethical — concerned with accuracy and reliability rather than fairness or social consequence.
The phrase AI auditing entered wider use in the early 2010s, initially in financial services and credit scoring contexts. Following the 2008 financial crisis, regulators and researchers began scrutinising the automated models that had shaped lending decisions, often to the systemic disadvantage of lower-income borrowers. This produced the first wave of structured thinking about what it would mean to independently examine an algorithmic system — not simply for technical performance but for discriminatory impact.
Fashion was not part of this early conversation. The industry’s adoption of AI was slower than financial services, and its accountability infrastructure was thinner. The term entered fashion discourse meaningfully only after 2016, when investigative reporting and academic research began documenting how automated hiring tools used by major retailers and logistics companies produced racially and gender-biased outcomes. Amazon’s 2018 disclosure that its internal AI recruiting tool had systematically downgraded CVs containing the word “women’s” became a widely cited turning point, shifting public understanding of algorithmic harm from abstract concern to documented reality.
By the late 2010s, the broader AI ethics movement — drawing on scholars including Joy Buolamwini, whose 2018 Gender Shades research exposed facial recognition bias, and Timnit Gebru, whose work on large language model risks brought internal industry accountability into public view — had created a framework within which AI auditing could be understood as a governance mechanism rather than a purely technical one.
In fashion specifically, the term gained operational relevance as brands began deploying AI in sourcing and supplier management. Tools that scored suppliers on risk, predicted order volumes, or automated compliance flags were making decisions with direct consequences for workers and producers in the Global South — yet no external body was examining them. Labour rights organisations began calling explicitly for algorithmic audits of these systems from approximately 2020 onwards.
The EU AI Act, formally adopted in 2024 and entering phased enforcement in 2026, represents the most significant regulatory moment in this history. For the first time, a major jurisdiction imposed legally binding audit requirements on high-risk AI systems, including those used in employment and procurement — categories directly relevant to fashion operations. The Act did not create an AI auditing profession overnight, but it created the legal mandate that now drives corporate compliance activity.
The term’s meaning has shifted across this period. Early usage emphasised technical accuracy: does the system do what it was built to do? Later usage incorporated fairness: does the system treat different groups equitably? Current usage has expanded further to include transparency, environmental impact, and governance: who built the system, on whose behalf, and to whose benefit?
AI auditing occupies an unusual cultural position: it is simultaneously a technical compliance function and a site of genuine political contestation. How this term is understood depends significantly on who is using it and in what context.
Within corporate environments, AI auditing is predominantly understood as a risk management exercise. The dominant concern is regulatory exposure — ensuring that deployed systems meet the minimum requirements of frameworks like the EU AI Act and do not generate reputational liability. In this context, auditing is a defensive instrument, and its scope tends to be defined by legal counsel rather than ethics practitioners. The audit is something that happens to a system before deployment, not an ongoing accountability mechanism.
Within civil society and labour rights communities, the term carries different weight. For organisations working on supply chain accountability — groups such as the Clean Clothes Campaign, the Business and Human Rights Resource Centre, and academic researchers in the algorithmic accountability field — AI auditing represents a potential tool for worker protection that is currently underused, underregulated, and frequently captured by the interests of the brands it is meant to scrutinise. The cultural resonance here is closer to financial auditing after Enron: a practice that looks rigorous but may be structurally compromised by its funding relationships.
In media coverage, AI auditing appears most frequently in one of two registers. The first is the corporate announcement — a brand or platform declares that its AI systems have been independently audited, typically without disclosing methodology, auditor identity, or findings. The second is the investigative exposure — a journalist or researcher demonstrates that a system widely described as fair or audited has produced documented harm. The gap between these two registers is culturally significant: it reflects a public that is increasingly sceptical of self-reported AI accountability.
Consumer perception of AI auditing is difficult to assess directly, because most consumers are unaware that the tools shaping their shopping experiences, the workers producing their garments, and the logistics infrastructure delivering their orders are AI-mediated. Awareness tends to emerge only through scandal. When an automated content moderation system suppresses a Black creator’s posts, or when a demand forecasting tool produces excess inventory that is incinerated, the systems themselves become visible — but the audit question rarely surfaces in mainstream coverage.
Regionally, attitudes diverge sharply. European markets, shaped by the GDPR era and now the EU AI Act, show greater regulatory and civil society engagement with AI auditing than markets in the United States, where federal AI legislation remains fragmented, or in Asia, where implementation varies significantly by jurisdiction. In sourcing countries — Bangladesh, Cambodia, Vietnam, Indonesia — the conversation about AI auditing is largely absent at the policy level, even as workers in these countries are among the most directly affected by algorithmic management tools deployed by the brands that source from them.
In fashion education, AI auditing is beginning to enter curricula at institutions with strong sustainability programmes, though coverage remains uneven. It sits at the intersection of data ethics, supply chain management, and sustainability governance — a combination that few courses currently integrate.
• Some AI auditing requires access to proprietary training datasets that companies refuse to disclose.
• AI systems used for demand forecasting can indirectly increase production volume if profit-maximisation logic dominates.
• There is no globally standardised sustainability metric for AI energy use yet.
ADVERT BOX
The concept of auditing algorithmic systems predates the term AI auditing itself. Its intellectual roots lie in software verification practices of the 1970s and 1980s, when computer scientists first grappled with how to validate that automated systems performed as specified. These early frameworks were technical, not ethical — concerned with accuracy and reliability rather than fairness or social consequence.
The phrase AI auditing entered wider use in the early 2010s, initially in financial services and credit scoring contexts. Following the 2008 financial crisis, regulators and researchers began scrutinising the automated models that had shaped lending decisions, often to the systemic disadvantage of lower-income borrowers. This produced the first wave of structured thinking about what it would mean to independently examine an algorithmic system — not simply for technical performance but for discriminatory impact.
Fashion was not part of this early conversation. The industry’s adoption of AI was slower than financial services, and its accountability infrastructure was thinner. The term entered fashion discourse meaningfully only after 2016, when investigative reporting and academic research began documenting how automated hiring tools used by major retailers and logistics companies produced racially and gender-biased outcomes. Amazon’s 2018 disclosure that its internal AI recruiting tool had systematically downgraded CVs containing the word “women’s” became a widely cited turning point, shifting public understanding of algorithmic harm from abstract concern to documented reality.
By the late 2010s, the broader AI ethics movement — drawing on scholars including Joy Buolamwini, whose 2018 Gender Shades research exposed facial recognition bias, and Timnit Gebru, whose work on large language model risks brought internal industry accountability into public view — had created a framework within which AI auditing could be understood as a governance mechanism rather than a purely technical one.
In fashion specifically, the term gained operational relevance as brands began deploying AI in sourcing and supplier management. Tools that scored suppliers on risk, predicted order volumes, or automated compliance flags were making decisions with direct consequences for workers and producers in the Global South — yet no external body was examining them. Labour rights organisations began calling explicitly for algorithmic audits of these systems from approximately 2020 onwards.
The EU AI Act, formally adopted in 2024 and entering phased enforcement in 2026, represents the most significant regulatory moment in this history. For the first time, a major jurisdiction imposed legally binding audit requirements on high-risk AI systems, including those used in employment and procurement — categories directly relevant to fashion operations. The Act did not create an AI auditing profession overnight, but it created the legal mandate that now drives corporate compliance activity.
The term’s meaning has shifted across this period. Early usage emphasised technical accuracy: does the system do what it was built to do? Later usage incorporated fairness: does the system treat different groups equitably? Current usage has expanded further to include transparency, environmental impact, and governance: who built the system, on whose behalf, and to whose benefit?
AI auditing occupies an unusual cultural position: it is simultaneously a technical compliance function and a site of genuine political contestation. How this term is understood depends significantly on who is using it and in what context.
Within corporate environments, AI auditing is predominantly understood as a risk management exercise. The dominant concern is regulatory exposure — ensuring that deployed systems meet the minimum requirements of frameworks like the EU AI Act and do not generate reputational liability. In this context, auditing is a defensive instrument, and its scope tends to be defined by legal counsel rather than ethics practitioners. The audit is something that happens to a system before deployment, not an ongoing accountability mechanism.
Within civil society and labour rights communities, the term carries different weight. For organisations working on supply chain accountability — groups such as the Clean Clothes Campaign, the Business and Human Rights Resource Centre, and academic researchers in the algorithmic accountability field — AI auditing represents a potential tool for worker protection that is currently underused, underregulated, and frequently captured by the interests of the brands it is meant to scrutinise. The cultural resonance here is closer to financial auditing after Enron: a practice that looks rigorous but may be structurally compromised by its funding relationships.
In media coverage, AI auditing appears most frequently in one of two registers. The first is the corporate announcement — a brand or platform declares that its AI systems have been independently audited, typically without disclosing methodology, auditor identity, or findings. The second is the investigative exposure — a journalist or researcher demonstrates that a system widely described as fair or audited has produced documented harm. The gap between these two registers is culturally significant: it reflects a public that is increasingly sceptical of self-reported AI accountability.
Consumer perception of AI auditing is difficult to assess directly, because most consumers are unaware that the tools shaping their shopping experiences, the workers producing their garments, and the logistics infrastructure delivering their orders are AI-mediated. Awareness tends to emerge only through scandal. When an automated content moderation system suppresses a Black creator’s posts, or when a demand forecasting tool produces excess inventory that is incinerated, the systems themselves become visible — but the audit question rarely surfaces in mainstream coverage.
Regionally, attitudes diverge sharply. European markets, shaped by the GDPR era and now the EU AI Act, show greater regulatory and civil society engagement with AI auditing than markets in the United States, where federal AI legislation remains fragmented, or in Asia, where implementation varies significantly by jurisdiction. In sourcing countries — Bangladesh, Cambodia, Vietnam, Indonesia — the conversation about AI auditing is largely absent at the policy level, even as workers in these countries are among the most directly affected by algorithmic management tools deployed by the brands that source from them.
In fashion education, AI auditing is beginning to enter curricula at institutions with strong sustainability programmes, though coverage remains uneven. It sits at the intersection of data ethics, supply chain management, and sustainability governance — a combination that few courses currently integrate.
• Some AI auditing requires access to proprietary training datasets that companies refuse to disclose.
• AI systems used for demand forecasting can indirectly increase production volume if profit-maximisation logic dominates.
• There is no globally standardised sustainability metric for AI energy use yet.
It’s a formal check-up for AI systems to make sure they work properly, don’t discriminate, don’t misuse data, and don’t create hidden environmental or governance risks.
2018–2019: The Hiring Bias Moment AI auditing gained mainstream attention in fashion’s periphery following Amazon’s disclosure that its internal recruiting algorithm systematically disadvantaged female candidates. While Amazon itself was not a fashion brand, the story landed in a sector heavily reliant on automated hiring for warehouse, logistics, and retail roles. The cultural shock of a major technology company acknowledging algorithmic discrimination prompted brands and retailers to begin — quietly — reviewing their own hiring tools. The trend was reactive rather than proactive, driven by reputational fear rather than governance conviction.
2020–2021: Supply Chain Scrutiny Intensifies The COVID-19 pandemic exposed the fragility and opacity of global fashion supply chains. As brands cancelled orders, withheld payments, and terminated supplier relationships through automated procurement systems, labour rights organisations began documenting the role of algorithmic tools in these decisions. This period saw the first explicit calls, from researchers and civil society groups, for independent audits of supplier-scoring and risk-flagging AI systems. The trend was driven by external pressure, not internal initiative.
2022–2023: Regulatory Anticipation As the EU AI Act moved through the legislative process, compliance-focused AI auditing activity increased substantially. Law firms, consultancies, and specialist auditing firms began building AI audit practices in anticipation of mandatory requirements. Fashion brands with significant European operations began conducting preliminary internal reviews. The trend was shaped almost entirely by regulatory trajectory rather than ethical leadership, and was concentrated in hiring and worker monitoring tools — the categories most clearly covered by high-risk classifications.
2024: Enforcement Infrastructure Builds Formal adoption of the EU AI Act in 2024 triggered a significant scaling of AI auditing activity across sectors, including fashion. Third-party auditing firms expanded capacity. Standards bodies including ISO and IEEE accelerated work on auditing methodologies. Simultaneously, academic research on auditing limitations — particularly the problem of auditing systems without full data access — gained prominence, creating pressure for more rigorous disclosure requirements. The trend in this period was characterised by rapid professionalisation alongside persistent methodological disagreement.
2025–2026: Scope Expands to Sustainability Growing pressure from CSRD reporting obligations and investor ESG requirements began connecting AI auditing to sustainability disclosure. Brands began receiving questions from institutional investors about whether their AI-driven sourcing and logistics tools had been assessed for environmental impact — specifically energy consumption and Scope 3 emissions contribution. This represented a meaningful expansion of the auditing concept beyond fairness and compliance into environmental accountability. The trend remains early-stage, with no standardised methodology yet established for sustainability-specific AI audits in fashion.
THE BASIC IDEA Independent or internal review of AI systems to verify they perform as claimed, without causing unintended harm or bias.
WHY THIS TERM EXISTS As AI shapes sourcing, hiring, and compliance decisions across fashion, the industry needed a mechanism to verify system performance and assign accountability when automated decisions cause harm.
SUSTAINABILITY STACK Primary: Labour, Power & Governance / Transparency & Traceability
AI auditing creates the accountability infrastructure that makes algorithmic decision-making legible, contestable, and governable.
CURRENT STATE OF DEVELOPMENT AI auditing in fashion remains nascent and largely voluntary. No industry-wide standard defines what a credible audit requires, who may conduct one, or how findings must be disclosed. Most activity concentrates in hiring and credit tools under regulatory pressure, leaving sourcing algorithms, demand forecasting models, and supplier-scoring systems almost entirely unexamined. Third-party auditing capacity is limited and methodology is inconsistent across practitioners.
COMMON MISUNDERSTANDINGS
✕ Auditing equals testing during development only
✕ Third-party audits guarantee system fairness
✕ Compliance audits and AI audits are interchangeable
✕ Auditing is a one-time certification event
✕ Open-source models require no auditing
BY THE NUMBERS
69% AUDIT GAP Only 69% of organisations deploying AI report conducting any form of internal audit before deployment.¹
$1.2tn BIAS COST Algorithmic bias in hiring and procurement decisions is estimated to cost the global economy $1.2 trillion annually.²
8% FASHION COVERAGE Fewer than 8% of fashion companies disclose any information about AI system performance or auditing processes.³
2026 ENFORCEMENT DATE The EU AI Act mandates conformity assessments for high-risk AI systems, with enforcement beginning August 2026.⁴
REGULATORY STATUS — 2026
| Regulation | Status | Description |
|---|---|---|
| EU AI Act | Enforced | High-risk AI systems in employment and supply chain contexts require documented conformity assessments before deployment. |
| EU CSRD | Enforced | Material AI-related risks affecting labour or emissions must be disclosed in sustainability reporting from financial year 2024. |
| NYC Local Law 144 | Enforced | Automated employment decision tools used in hiring must undergo annual bias audits and publish results publicly. |
THE HONEST TENSION AI auditing is structurally limited by who commissions it. Brand-funded audits create incentive to scope narrowly, publish selectively, and treat findings as proprietary. Truly independent auditing requires data access that most companies will not grant and technical capacity that most regulators do not yet hold. The audit as currently practised risks becoming a legitimacy signal rather than a genuine accountability mechanism — greenwashing’s algorithmic equivalent.
RISKS AND UNINTENDED CONSEQUENCES Auditing can create false assurance if scope is insufficient or methodology is undisclosed. A system certified as fair under one demographic dataset may perform poorly across others. Audit findings treated as confidential undermine the transparency the process is intended to produce. In fashion supply chains, where audited brands hold power over suppliers, audit results can be weaponised to justify terminating relationships rather than remedying system failures.
KNOWLEDGE GAPS No consensus exists on what constitutes a complete AI audit. Metrics for fairness conflict across methodologies. Fashion-specific auditing frameworks covering sourcing, forecasting, and pricing algorithms do not yet exist.
Books
- Atlas of AI — Kate Crawford
- Artificial Intelligence: A Guide for Thinking Humans — Melanie Mitchell
- The Alignment Problem — Brian Christian
References
- ¹ KPMG. (2023). Trust in artificial intelligence: A global study. KPMG International.
- ² National Bureau of Economic Research. (2022). The economic costs of algorithmic bias in labour and procurement markets. NBER Working Paper No. 29875.
- ³ Business of Fashion & McKinsey & Company. (2024). The state of fashion: Technology. BoF Insights.
- ⁴ European Parliament. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union.
- Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81, 1–15.
- Brundage, M., Avin, S., Clark, J., Toner, H., Eckersley, P., Garfinkel, B., Dafoe, A., Scharre, P., Zeitzoff, T., Filar, B., Anderson, H., Roff, H., Allen, G. C., Steinhardt, J., Flynn, C., hÉigeartaigh, S., Beard, S., Belfield, H., Farquhar, S., … Amodei, D. (2020). Toward trustworthy AI development: Mechanisms for supporting verifiable claims. arXiv preprint arXiv:2004.07213.
- European Commission. (2020). White Paper on Artificial Intelligence: A European approach to excellence and trust.
- European Commission. (2021). Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) (COM/2021/206 final).
- European Parliament & Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act).
- European Financial Reporting Advisory Group (EFRAG). (2023). European Sustainability Reporting Standards (ESRS).
- International Energy Agency. (2022). Data centres and data transmission networks.
- International Energy Agency. (2023). Electricity Market Report 2023.
- ISSB. (2023). IFRS S1 General Requirements for Disclosure of Sustainability-related Financial Information and IFRS S2 Climate-related Disclosures.
- Kroll, J. A., Huey, J., Barocas, S., Felten, E. W., Reidenberg, J. R., Robinson, D. G., & Yu, H. (2017). Accountable algorithms. University of Pennsylvania Law Review, 165(3), 633–705.
- Li, Y., Wen, Y., Tao, D., & Guan, K. (2023). Making AI less thirsty: Uncovering and addressing the secret water footprint of AI models. arXiv preprint arXiv:2304.03271.
- Mytton, D. (2021). Data centre water consumption. Nature Computational Science, 1, 471–472.
- NIST. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology.
- OECD. (2019). OECD Principles on Artificial Intelligence.
- OECD. (2023). OECD AI Policy Observatory.
- O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown.
- Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L., Rothchild, D., So, D., Texier, M., & Dean, J. (2021). Carbon emissions and large neural network training. arXiv preprint arXiv:2104.10350.
- Raji, I. D., Smart, A., White, R. N., Mitchell, M., Gebru, T., Hutchinson, B., Smith-Loud, J., Theron, D., & Barnes, P. (2020). Closing the AI accountability gap: Defining an end-to-end framework for internal algorithmic auditing. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 33–44.
- Samuelson, P. (2023). Generative AI meets copyright. Columbia Journal of Law & the Arts.
- Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 3645–3650.
- U.S. Copyright Office. (2023). Copyright registration guidance: Works containing material generated by artificial intelligence.
- UK Intellectual Property Office. (2022). Artificial intelligence and intellectual property: Copyright and patents consultation response.
- World Economic Forum. (2022). Toolkit for Board Governance of Artificial Intelligence.
- World Economic Forum. (2023). The Global Risks Report 2023.
Related Articles
Fashion in the Regency Era, (1811–1820), nestled within the broader...
Fashion Accountability Report: Bridging the Gap Between Promise and Progress...