Definition
A structured way of classifying the different kinds of weaknesses that show up in AI systems — separating, for example, flaws in the model itself from flaws in the surrounding software (like tool connectors or dashboards). It gives security teams a shared map so they can decide where to focus testing and patching effort, instead of chasing each new attack technique one at a time.
Why it matters
Without a common map of 'what can go wrong and where,' security budgets and audits tend to chase headlines instead of systemic risk; a taxonomy helps boards ask sharper questions about coverage gaps.