The article explores why Ontario Digital Service could not procure '98% safe' Large Language Models (LLMs) for public sector use. The core issue lies in the inherent risks of generative AI, where absolute safety is technically unachievable due to potential hallucinations and data leakage. This procurement hurdle highlights the critical challenge of balancing innovation with data security and privacy compliance in government operations.

The implications are significant for Ontario's 15 million residents, as it delays the integration of potentially efficiency-boosting AI tools into public services. The analysis suggests that current vendor claims of safety are insufficient for rigorous public sector standards. This situation serves as a case study for other jurisdictions grappling with AI adoption.

It underscores the need for robust governance frameworks before deploying LLMs in sensitive environments. The Ontario Digital Service must navigate these technical limitations while ensuring public trust and safety.