When AI policy hallucinates

South Africa’s withdrawn draft National AI Policy highlights a critical lesson about trust and accountability in AI governance - a perspective recently shared by Fatima Ismail, Senior Associate in the IPTech Practice Group, during a legal discussion on Voice of Wits (VoW 88.1 FM). While the draft policy was intended as a roadmap for future regulation, its withdrawal following reportedly fictitious references underscores the importance of credible, well-verified policymaking processes. Beyond citation errors, the incident raises broader concerns about how to responsibly govern AI in a context where outputs can appear authoritative yet be incorrect. This is particularly significant given AI’s growing role across sectors such as finance, healthcare, and public services, where risks differ and higher-stakes applications require stronger safeguards, including transparency, oversight and accountability. It also highlights that AI regulation will likely span multiple sectors, supported by existing frameworks like POPIA. The key takeaway is that while AI can enhance and support decision-making, it cannot replace human accountability - and South Africa’s path forward will require a balanced, risk-based approach that fosters innovation while protecting individuals from harm.