AI Governance Engineer
GitHub

Guardrail

By Updated

A runtime control that inspects or mediates a model's or agent's inputs, outputs or tool calls and blocks, rewrites or escalates what breaks a policy, logging each decision as evidence. Guardrails are deterministic code or classifiers in the call path, unlike a guardian agent, which is itself an AI system.

Developed in
ch. 05, Pattern: Runtime Guardrail
ch. 23, Runtime guardrails for tool calls
Chapters
ch. 04, The Stack · ch. 05, Patterns · ch. 23, AI Agents
Contrast with
Guardian agent
Source
Defined by this Body of Knowledge: the section it is developed in is the source

Where it is used

21 chapters of the Body of Knowledge use the term. Each link opens the first section that does.

Guardrail: mentions per chapter140 mentions of the term across 21 of the 23 chapters of the Body of Knowledge; the chapters the glossary cross-references are drawn solid.Cross-referenced chapterOther chapter01 Definition: 3 mentions01 Definition302 Why Now: 1 mention02 Why Now103 Values & Principles: 5 mentions03 Values & Principles504 The Stack: 15 mentions04 The Stack1505 Patterns: 4 mentions05 Patterns406 The Role: 3 mentions06 The Role307 Maturity Model: 5 mentions07 Maturity Model508 Regulatory Map: 9 mentions08 Regulatory Map910 Reading List: 2 mentions10 Reading List211 AI Defined: 15 mentions11 AI Defined1512 Governance Program: 4 mentions12 Governance Program413 Risk Management: 13 mentions13 Risk Management1314 Development: 2 mentions14 Development215 Deployment: 14 mentions15 Deployment1417 Incidents: 18 mentions17 Incidents1818 EU AI Act: 1 mention18 EU AI Act119 Privacy & AI: 3 mentions19 Privacy & AI320 Existing Law: 6 mentions20 Existing Law621 AI Laws Worldwide: 1 mention21 AI Laws Worldwide122 Principles & Standards: 3 mentions22 Principles & Standards323 AI Agents: 13 mentions23 AI Agents13No mentions in the other 2 chaptersSource: the Body of Knowledgev0.5.0
Guardrail: mentions per chapter140 mentions of the term across 21 of the 23 chapters of the Body of Knowledge; the chapters the glossary cross-references are drawn solid.Cross-referenced chapterOther chapter0001 Definition: 3 mentions30102 Why Now: 1 mention10203 Values & Principles: 5 mentions50304 The Stack: 15 mentions150405 Patterns: 4 mentions40506 The Role: 3 mentions30607 Maturity Model: 5 mentions50708 Regulatory Map: 9 mentions90810 Reading List: 2 mentions21011 AI Defined: 15 mentions151112 Governance Program: 4 mentions41213 Risk Management: 13 mentions131314 Development: 2 mentions21415 Deployment: 14 mentions14151617 Incidents: 18 mentions181718 EU AI Act: 1 mention11819 Privacy & AI: 3 mentions31920 Existing Law: 6 mentions62021 AI Laws Worldwide: 1 mention12122 Principles & Standards: 3 mentions32223 AI Agents: 13 mentions1323Source: the Body of Knowledge v0.5.0
Mentions per chapter How often each chapter names the term, with the chapters the glossary cross-references drawn solid. The longest bar is where to read the term in use; bars spread across many chapters mark a term that runs through the discipline.
Mentions as a table
Mentions of Guardrail per chapter
ChapterMentionsCross-referenced
00 Preface0No
01 Definition3No
02 Why Now1No
03 Values & Principles5No
04 The Stack15Yes
05 Patterns4Yes
06 The Role3No
07 Maturity Model5No
08 Regulatory Map9No
10 Reading List2No
11 AI Defined15No
12 Governance Program4No
13 Risk Management13No
14 Development2No
15 Deployment14No
16 Fairness & XAI0No
17 Incidents18No
18 EU AI Act1No
19 Privacy & AI3No
20 Existing Law6No
21 AI Laws Worldwide1No
22 Principles & Standards3No
23 AI Agents13Yes

Patterns that use this term

12 pattern pages use the term; the 10 that use it most:

Guardrail: related terms and patternsThe term at the centre; related terms on solid edges, the terms it is contrasted with on dashed edges, and the patterns that use it most in the colour of their layer.RelatedContrastColour: stack layer of the pattern01 Govern-as-Code04 Runtime Controls05 AssuranceGuardrailRelated terms (7)Design defect · RelatedDesign defectFailure posture · RelatedFailure postureManufacturing defect · RelatedManufacturing defectMultimodal model · RelatedMultimodal modelPrompt injection · RelatedPrompt injectionSmall language model (SLM) · RelatedSmall language model (SLM)Version pinning · RelatedVersion pinningContrast with (1)Guardian agent · ContrastGuardian agentPatterns using it most (6)AI Threat Model, 5 mentions · Uses the termAI Threat ModelContinuous Assurance Telemetry, 4 mentions · Uses the termContinuous Assurance TelemetryFRIA-as-Code, 3 mentions · Uses the termFRIA-as-CodeHuman-in-the-loop Gate, 3 mentions · Uses the termHuman-in-the-loop GateIncident Pipeline, 3 mentions · Uses the termIncident PipelineRuntime Guardrail, 18 mentions · Uses the termRuntime Guardrail
Guardrail: related terms and patternsThe term at the centre; related terms on solid edges, the terms it is contrasted with on dashed edges, and the patterns that use it most in the colour of their layer.RelatedContrastColour: stack layer of the pattern01 Govern-as-Code04 Runtime Controls05 AssuranceRelated terms (7)Design defect · RelatedDesign defectFailure posture · RelatedFailure postureManufacturing defect · RelatedManufacturing defectMultimodal model · RelatedMultimodal modelPrompt injection · RelatedPrompt injectionSmall language model (SLM) · RelatedSmall language model (SLM)Version pinning · RelatedVersion pinningContrast with (1)Guardian agent · ContrastGuardian agentPatterns using it most (6)AI Threat Model, 5 mentions · Uses the termAI Threat ModelContinuous Assurance Telemetry, 4 mentions · Uses the termContinuous Assurance TelemetryFRIA-as-Code, 3 mentions · Uses the termFRIA-as-CodeHuman-in-the-loop Gate, 3 mentions · Uses the termHuman-in-the-loop GateIncident Pipeline, 3 mentions · Uses the termIncident PipelineRuntime Guardrail, 18 mentions · Uses the termRuntime GuardrailGuardrail
The term's neighbourhood Terms related to this one on solid edges, terms it is often confused with on dashed edges, and the patterns that use it most in the colour of their layer. Follow a dashed edge to check the distinction before you use either term.
Neighbourhood as a table
Neighbourhood of Guardrail
FamilyItemRelation
Related termsDesign defectRelated
Related termsFailure postureRelated
Related termsManufacturing defectRelated
Related termsMultimodal modelRelated
Related termsPrompt injectionRelated
Related termsSmall language model (SLM)Related
Related termsVersion pinningRelated
Contrast withGuardian agentContrast
Patterns using it mostAI Threat Model, 5 mentionsUses the term
Patterns using it mostContinuous Assurance Telemetry, 4 mentionsUses the term
Patterns using it mostFRIA-as-Code, 3 mentionsUses the term
Patterns using it mostHuman-in-the-loop Gate, 3 mentionsUses the term
Patterns using it mostIncident Pipeline, 3 mentionsUses the term
Patterns using it mostRuntime Guardrail, 18 mentionsUses the term
  • Failure posture

    What a guardrail, guardian agent or tool gateway does when it cannot reach a decision: fail open lets the call through, fail closed blocks it.

  • Small language model (SLM)

    A language model small enough to run close to the user, for example on a phone.

  • Manufacturing defect

    In product liability, a departure of a unit from its own design.

  • Multimodal model

    A model that takes or produces more than one modality (text, image, audio, video).

  • Prompt injection

    An input that alters a model's behaviour or output in ways its designers did not intend.

  • Version pinning

    Fixing, in the registry entry, the exact versions of the model, prompts, retrieval corpus and guardrails a deployed system uses, so what ran is known and any…

  • Design defect

    In product liability, a defect inherent in the design of every unit, judged by consumer expectations or by weighing risk against utility.

Sources

No external source: the term is coined or used in a specific sense by this Body of Knowledge, and the section linked under "Developed in" is its source.

Definitions of legal terms paraphrase the cited text, which governs. Dated statements are as of .

Cite this term

García Aibar, J. (2026). Guardrail. In AI Governance Engineering: The Thesis & Body of Knowledge (v0.5.0), Glossary. https://doi.org/10.5281/zenodo.22956197. https://aigovernanceengineer.com/glossary/guardrail. CC BY 4.0

BibTeX

@misc{aige2026guardrail,
  author  = {Jorge García Aibar},
  title   = {{Guardrail}},
  note    = {Glossary, AI Governance Engineering: The Thesis \& Body of Knowledge, version 0.5.0},
  year    = {2026},
  doi     = {10.5281/zenodo.22956197},
  url     = {https://aigovernanceengineer.com/glossary/guardrail}
}