Eticas study: Global South AI auditing nearly nonexistent with under 20 audits
AI is deployed everywhere in the Global South, but almost nobody is checking its impact.
A new working paper from Eticas Foundation, led by Gemma Galdon Clavell and Alexandra Magaard, documents the sparse state of algorithmic auditing in the Global South. Over a decade of practice including the only fully published second-party audit (Robot Laura in Brazil), two completed but unreleased national audits, and thirteen Responsible AI Assessments, the researchers counted fewer than twenty published audits of deployed systems. This stands against hundreds of documented public-sector algorithms and multibillion-dollar national AI investments across Latin America, Sub-Saharan Africa, and Asia Pacific.
The paper identifies four cross-cutting patterns that undermine AI accountability: proxy targets that substitute predictability for validity, performance claims that collapse under prevalence analysis, populations scored by models that never saw them in training, and structural bias persisting even after removing protected attributes. The authors argue the root cause is not lack of technical capacity but lack of funded demand—no actor is required or funded to hold deployed systems to account. They recommend that development and philanthropic funders, who finance most consequential AI in the region, require independent evaluation as a condition of funding.
- Fewer than 20 published second- and third-party audits of deployed AI systems in the Global South over the past decade
- Four systemic patterns: proxy targets, collapsing performance claims, missing training data, and persistent structural bias
- Root cause identified as funding gap, not capacity gap—funders must mandate independent evaluation
Why It Matters
Without auditing, AI deployed in Global South risks replicating bias and harm at scale, unchecked by regulation.