Glossary and Unified Taxonomy¶
Concepts like PACE, MASO, and risk tiers span both this site and airuntimesecurity.io. To keep a single source of truth, the canonical definitions of the shared runtime taxonomy live on airuntimesecurity.io and this page links straight to them. Pre-runtime terms specific to Secure by Design are defined here.
Why two glossaries, one source
Shared terms are defined once, on the runtime site, and linked by anchor. Where a term below points to airuntimesecurity.io/glossary/#..., that is the authoritative definition. This site only defines the pre-runtime terms it owns.
Shared taxonomy (canonical on airuntimesecurity.io)¶
| Term | Definition |
|---|---|
| PACE | Primary, Alternate, Contingency, Emergency resilience model. → canonical |
| MASO | Multi-Agent Security Operations. → canonical |
| ASO | Agent Security Operations (single-system AIRS). → canonical |
| Risk tiers | LOW / MEDIUM / HIGH / CRITICAL, and the simplified Tier 1/2/3 shorthand. → canonical · this site's tiers |
| Guardrails (Layer 01) | Deterministic prevention. → canonical |
| Model-as-Judge (Layer 02) | Probabilistic detection. → canonical |
| Human Oversight (Layer 03) | Human decision on high-consequence actions. → canonical |
| Circuit breaker | Containment failsafe, maps to PACE Emergency. → canonical |
| Objective Intent | An agent's declared, checkable objective. → canonical |
Pre-runtime taxonomy (canonical here)¶
Secure by Design¶
AI security before deployment: model selection, platform selection, DevOps, MLOps, data security, and governance. This site decides what to enforce and builds the infrastructure that makes enforcement possible at runtime. Its runtime counterpart is AI Runtime Security. The point at which the two meet is the Control Handoff Matrix.
AI-aware SDLC¶
The software development lifecycle adapted for AI systems: risk classification, threat modelling, secure design, adversarial testing, and production-readiness gates, in addition to the normal SDLC. See AI-Aware SDLC.
Model provenance and integrity¶
Evidence of where a model came from and proof that its weights have not been tampered with: signed artefacts, hash verification, and a documented chain of custody before a model is allowed into a pipeline (SUP-01). See Provenance and Integrity.
Model risk assessment¶
The structured evaluation of a model's suitability and risk before adoption, feeding the tier decision and the guardrail configuration it will need (SUP-02). See Model Risk Assessment.
Infrastructure controls (the 80)¶
The 80 technical controls that make behavioural security enforceable, across identity, logging, network, data, secrets, supply chain, incident response, and four agentic domains. This site mirrors them into the pre-runtime lifecycle; airuntimesecurity.io is authoritative for their definitions. See the Infrastructure Controls and the machine-readable catalog.
Control ID scheme¶
Every infrastructure control has a stable ID of the form PREFIX-NN:
| Prefix | Domain | Prefix | Domain |
|---|---|---|---|
IAM |
Identity & Access | IR |
Incident Response |
LOG |
Logging & Observability | TOOL |
Agentic - Tool Access |
NET |
Network & Segmentation | SESS |
Agentic - Session & Scope |
DAT |
Data Protection | DEL |
Agentic - Delegation Chains |
SEC |
Secrets & Credentials | SAND |
Agentic - Sandbox Patterns |
SUP |
Supply Chain |
AI-BOM¶
AI Bill of Materials. A maintained inventory of the models, datasets, tools, and dependencies a system uses, so coverage and provenance can be verified systematically (SUP-07). The AI analogue of a software SBOM.