Closing the AI Assurance Divide

Key Takeaways

• AI assurance — the process of assessing and communicating AI trustworthiness — offers a path to balance AI innovation with safety, security, and fairness. But it has developed primarily in advanced economies, leaving many AI risks unaddressed in developing countries that represent some of the world’s largest AI markets.
• Closing the AI assurance divide can advance the Sustainable Development Goals, fulfill states’ human rights obligations and companies’ corporate responsibilities, and ensure trust in AI is warranted where it exists, while building justified confidence where it’s lacking.
• Building effective AI assurance ecosystems in developing countries faces acute challenges, including vast diversity across and within countries, resource-intensive requirements for specialized skills and infrastructure, and legal and political hurdles such as regulatory gaps and limited international representation.
• Developing nations need flexible strategies that balance technical rigor, socio-technical sensitivity, and cost-effectiveness rather than a one-size-fits-all approach.
• Five overarching strategies can help lay the foundations of robust national AI assurance ecosystems: selecting what to assure, understanding how to assure, building the capacity to assure, setting assurance criteria, and cooperating internationally.

Summary

AI assurance is critical for balancing innovation with safety across the world’s AI markets. However, this field has developed primarily in advanced economies, creating an assurance divide that leaves developing countries vulnerable despite representing some of the largest AI markets globally.

This brief examines the challenges of building AI assurance ecosystems in developing countries and proposes practical strategies for bridging the global AI assurance divide. AI risks are often amplified in the Global South, where levels of trust in AI vary from over-reliance to skepticism. Robust assurance can ensure warranted trust and build justified confidence while advancing sustainable development and human rights.

Building effective assurance ecosystems faces acute obstacles, including vast diversity across languages, cultural values, and risk profiles; resource constraints affecting infrastructure and specialized skills; and legal and political hurdles from regulatory gaps to limited international representation.

Strategies for building robust AI assurance ecosystems

1. Selecting what to assure, grounded in national AI risk assessments and taxonomies.
2. Understanding how to assure, by choosing context-appropriate assurance tools and techniques, such as evaluations, auditing, and certification.
3. Building the capacity to assure, through investments in AI literacy, infrastructure, skills development, and sustainable public and private funding.
4. Setting assurance criteria, using a mix of rules, norms, and/or standards that balance specificity with flexibility.
5. Cooperating internationally, through North–South, South–South, and multilateral partnerships that enable knowledge sharing, pooled resources, and stronger representation of Global South perspectives.

Looking ahead, this brief highlights key takeaways and priorities for future work. These include deepening socio-technical approaches to AI assurance; identifying a global baseline of assurance tools and techniques; determining where significant local adaptation or entirely new tools are required; prioritizing high-impact use cases and low-hanging fruit, such as organizational assurance; and addressing internal divides within countries, including socio-economic inequalities and gaps between public expectations and government action.