Tax, trust, and the geopolitics of AI infrastructure

The South African Revenue Service (SARS) has embedded artificial intelligence (AI) and machine learning (systems that learn patterns from data) across its tax administration, from automatically assessing returns to selecting which cases warrant a closer audit, TechCabal reports. SARS told TechCabal that AI-assisted fraud detection prevented more than R417 billion in improper refund payouts over five years, and that automated risk tools now select 100% of verification cases and nearly 89% of complex audit cases. As digital income from e-commerce, freelancing and cryptocurrency grows, the shift means more South Africans' financial data is being cross-referenced at a scale no manual process could match, raising questions about algorithmic accountability that SARS says it addresses through human oversight of every AI-generated risk flag.

US Treasury Secretary Scott Bessent has warned that sanctions and Entity List designations (a US government blacklist that restricts American companies from doing business with named foreign firms) are now being considered against Chinese AI companies, according to TechCentral. The move is part of escalating US-China tensions over the global AI industry. For South Africa, the practical effect could include reduced access to certain Chinese AI tools or hardware, and pressure on local companies to navigate an increasingly divided technology landscape.

Business Day reports that South African company executives face growing personal legal liability as AI tools make cyber attacks faster and harder to defend against. The podcast discussion centres on how boards are being held accountable for data breaches and whether current governance structures are equipped to manage AI-driven security risks. For South African organisations, the message is that cybersecurity is shifting from a technical concern to a boardroom responsibility with direct legal consequences.

Huawei South Africa's annual Connect event in Sandton highlighted deepening technology ties between South Africa and China, with CEO Will Meng noting that South Africa is a founding member of the newly launched World Artificial Intelligence Cooperation Organisation, established at the World Artificial Intelligence Conference in Shanghai. Meng argued that China's path of building digital infrastructure before deploying AI applications offers a relevant model for South Africa, and outlined Huawei's role in providing the networks, data centres, and computing power that local AI adoption would require. The event points to a practical question for South African organisations: as the cost of running AI rises, Chinese large language models (AI systems trained on vast text data to generate and process language) such as DeepSeek and Qwen, running on Huawei infrastructure, may become an increasingly considered alternative to offerings from Western technology companies.

Cybersecurity firm Kaspersky warns that attackers are using AI agents (software that can take actions autonomously on a user's behalf) to conceal malicious code in ways that traditional antivirus tools cannot detect, according to TechCentral. The firm says behaviour-based detection tools offer some protection, but that new defences are needed to counter techniques that exploit how AI agents process instructions. For South African organisations and individuals adopting AI-enabled tools, the finding points to a gap between the pace of AI adoption and the security measures currently in place.