LIRNEasia CEO Helani Galpaya participated as a speaker at the Small Island Developing States Supreme Audit Institutions Symposium (SIDS SAIs), held in Ukulhas, Maldives, from 3–5 August 2026. The symposium was organised by the Auditor General’s Office of the Maldives and the United Nations Department of Economic and Social Affairs (UNDESA) and brought together over 50 participants from 26 countries. Speakers over the three-day event included Dr. Mahadeva Muizzu (President of the Maldives), Mr. Hussein Niyazy (Auditor General of the Republic of the Maldives), and several other government ministers. 

Helani participated in the session “Digital Transformation and AI in Public Sector Governance in SIDs”. The session opened with a presentation by Dr. Mohamed Kinaanath (Minister of State for Homeland Security, Labour and Technology of the Maldives), Mr. Guilherme Canala (Director of the Digital Inclusion and Digital Transformation Division at UNESCO), and Mr. Stephen Sanford (Director at the US Government Accountability Office). This was followed by a fireside chat in which Helani and Dr. Mohamed Shareef (Former Minister of State for Environment, Climate Change and Technology of the Maldives) engaged in a discussion with the moderator, Ms. Archana Shirsat, Deputy Director General, INTOSAI. 

Helani covered several points on the opportunities and challenges associated with AI. 

Supply and Demand Challenges 

Helani highlighted the need to ensure that a new AI divide (similar to the digital divide) doesn’t emerge, noting that we still have to work on access, connectivity, and cost. For example, she explained that data costs are low in many South Asian countries, though for the lowest-income deciles they don’t meet the globally agreed-upon standard of costs being under 2% of monthly income. However, the AI subscription fees, which are around USD 20 and are not adjusted for purchasing power, make them quite expensive. 

Helani also noted the importance of how governments price digitalised services. The right incentives can ensure that the majority of people move to digitally provided services. She explained that the long-term cost savings from lower transaction costs enabled through digitalisation can fund the costlier service provision for the smaller number of citizens who are unable to use the digital version for various reasons. 

On the other hand, she discussed that AI applications, when done right, are more likely to help those who are less literate and may benefit from voice-based interactions with information, making information more accessible.  Helani also emphasised that the government’s role in digitalisation is to supply the minimum set of digital goods and services that are non-rivalrous and non-excludable, which enable others (the private sector, civil society and various government departments) to build upon those services. 

Governance, Accountability and Harms 

On the implications of AI governance, Helani emphasised that accountability and the related governance structures are as vital as they have ever been. She noted the need to avoid the classic problem of a poor worker being told that the “machine” denied his/her application for social welfare, without even being offered the opportunity for human review. Taking this to the larger issues of liability and responsibility, she pointed out that we are still far from having the AI governance we need to ensure that equitable development is enabled. 

Helani noted that when we talk about “ethical AI” and “responsible AI”, and similar terms, various guidelines define these terms slightly differently. She suggested that it might be useful to ask what exactly we are trying to protect. She mentioned that, for this, it is worth looking at a set of principles that have the widest adoption by nations – that is, the protection and enhancement of socio-economic rights, as well as human rights, as defined by UN conventions. 

On human rights, Helani points out that this part is reasonably well understood – violations of privacy through personal data use, loss of democracy through the spread of misinformation, and many other examples are studied, though we don’t yet have perfect solutions. On socio-economic rights, Helani noted that these are now largely in focus. At an individual level, this could include being left out of financial services based on social media scoring; but, more indirectly, it could include the costs of huge AI investments and the environmental harms that are borne by taxpayers, or situations where the resulting efficiency gains from AI are not enough to compensate for massive job losses.  

She also highlighted how dependence on a handful of global AI firms increases the risks for everyone else. She warned that geopolitical risks could mean a country suddenly gets banned from accessing models that are mission critical. Developing local models that can handle local languages and local context may be important to manage this risk. 

The role of Supreme Audit Institutions (SAIs) in auditing AI models 

Helani discussed how Supreme Audit Institutions in Small Island Developing States are currently not well placed to audit AI models and applications used by other government institutions. In fact, very few organisations, even in developed countries, have the capabilities to conduct these kinds of audits. 

She suggested that a better approach may be for SAIs to start using AI. Current levels of AI are well suited to improving the performance of SAIs, particularly as AI is great at pattern recognition and outlier identification – two key functions auditors need to identify fraudulent transactions. AI is also well suited to handling very large data sets, so instead of sampling some transactions and tracing them through, auditors can trace all transactions. Furthermore, they can go down to the level of a request for proposal, contract, terms of reference and other supporting documents in the case of government procurement, because AI is great at summarising large sets of documents and identifying themes. The multimodal capabilities of AI (e.g. analysing images) can also be used to conduct other audits, such as checking if the road was actually built. 

Helani explained that using AI for a while will give SAIs a deeper understanding of how AI can and cannot be audited, as well as the type of skills that are needed. The SAIs are then ready to work with AI Safety Institutes and others to set up mechanisms across all parts of the AI development chain, from design onwards. These could include things like developing toolkits that enable developers to test their models and applications (the SAIs develop the tools needed for anti-corruption and audits; safety institutions develop the tools for responsible AI principles; and so on).