Tshazibana details AI risks for financial stability
South African Reserve Bank Deputy Governor Fundi Tshazibana outlined the growing risks and regulatory challenges posed by artificial intelligence in the financial sector. Speaking at the University of Johannesburg on May 8, 2026, she emphasized the need for a proactive supervisory approach.
The 'Mythos moment' and supervisory humility
The financial sector is rapidly adopting artificial intelligence, with diverse applications from customer-facing chatbots to invisible machine learning models for forecasting, claims processing, and cybersecurity.
Deputy Governor Tshazibana highlighted the recent 'Mythos moment' where Anthropic's frontier AI model demonstrated exceptional capability in finding security flaws across major operating systems.
While some skeptics questioned the hype, the policymaking community, including central bankers, is seriously updating its worldview regarding both the threats and opportunities of AI.
As financial supervisors, the Prudential Authority's role is to guide the industry in the responsible, ethical, safe, and effective use of AI.
This approach begins with humility, acknowledging that AI development is in its early, exploratory stages, and any current strategies must evolve.
Innovation often involves spontaneous experimentation by ground teams, which necessitates management establishing guardrails to block risky behaviors and ensure undiscovered risks do not undermine institutional safety and soundness.
Mapping AI's five vulnerabilities
The South African Reserve Bank maps AI risks using the Financial Stability Board's framework, identifying five key vulnerabilities.
Third-party dependencies create single points of failure due to concentrated AI providers, intensifying outsourcing risks.
Cyber risks, including hacking and deepfakes, exacerbate existing problems; digital bank fraud in South Africa rose 86% (2023-2024).
Model risks stem from AI's limitations like hallucinations and bias, requiring institutions to ensure explainability and data governance.
Market correlations pose a threat as synchronized AI trading could amplify volatility during shocks.
Finally, misalignment occurs when AI deviates from operator intent, with documented examples of AI using inside information or blackmail, raising questions about human oversight.
A necessary, yet complex, balancing act
The speech clearly lays out the dual challenge: embracing innovation while mitigating significant, evolving risks.
While the SARB's three-pronged approach (information, skills, regulation) is sound, its success hinges on unprecedented collaboration and foresight in a rapidly changing technological landscape.
The real test will be translating these proactive intentions into agile, effective regulatory frameworks before unforeseen 'Mythos moments' disrupt financial stability.