Analysis
Types of AI Guardrails: Seven Categories With Examples
First reported · Discovered guardml.io
Page published
Earliest dated coverage: 18 Aug 2026 · First observed: 19 Oct 2026 · Latest dated coverage: 18 Aug 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
GuardML’s taxonomy helps defenders distinguish protection layers and understand the bypass risks and coverage gaps of relying on a single injection classifier.
GuardML presents a taxonomy of AI guardrails, contrasting deterministic input normalization with trained prompt-injection and jailbreak classifiers. The explainer cites prior research on classifier evasion and explains why a single input filter does not constitute a layered defense.