Generative AI has shifted from a novelty to standard operational tooling almost overnight, and the resulting governance gap is now surfacing in real-world incidents. The Deloitte Australia matter provides a valuable lens for any advisory practice implementing AI-assisted drafting, research, or coding.
The fundamental lesson is not to avoid AI, but to design the control environment first. This requires documented review checkpoints, clear provenance tracking for AI-assisted outputs, and defined escalation pathways when confidence is low.
We frame AI adoption around three essential controls: human-in-the-loop review for any client-facing output, an AI usage policy signed by every team member, and periodic sampling of AI-assisted work to catch drift before it becomes a finding.