# AI Governance Glossary

## EU AI Act
European Union regulation classifying AI systems by risk level: minimal, limited, high, unacceptable. Sets obligations for providers and deployers.

## BIO2
Baseline Informatiebeveiliging Overheid — Dutch security framework for government organizations based on ISO 27001/27002.

## NIS2
Network and Information Security Directive 2 — EU directive setting cybersecurity requirements for essential and important entities.

## NORA
Nederlandse Overheid Referentie Architectuur — reference architecture framework for Dutch government ICT.

## Cloud Sovereignty
Control over data, technology, operation and dependencies, including jurisdiction, ownership, access, keys and exit options. Hosting location alone does not guarantee immunity from foreign access.

## Zero Trust
Security paradigm where no user, device, or network segment is implicitly trusted; every access decision requires explicit verification.

## Agentic AI
AI systems that autonomously plan, execute, and iterate tasks with minimal human intervention, including tool use and multi-step reasoning.

## RAG
Retrieval-Augmented Generation — architecture enriching LLMs with relevant context from external knowledge bases before generating responses.

## Red Teaming
Structured approach where a team attempts to break an AI system's defenses to identify weaknesses before malicious actors do.

## Prompt Injection
Attack technique embedding malicious instructions into LLM input to manipulate the model outside its intended scope.

## Secure SDLC
Security-integrated Software Development Lifecycle embedding security practices at every phase: plan, code, build, test, release, deploy, operate, monitor.

## Supply Chain Security
Security measures to minimize risks in the procurement chain of software, hardware, and services.

## FinOps
Financial Operations practice to optimize cloud costs through collaboration between finance, engineering, and business teams.

## DPIA
Data Protection Impact Assessment — mandatory privacy impact analysis for high-risk data processing under GDPR.

## Common Ground
Dutch standardization program promoting interoperability and efficiency in government ICT.

## DevSecOps
Development methodology integrating security into every phase of the software development lifecycle.

## SRE
Site Reliability Engineering — applying software engineering practices to infrastructure and operations for reliable, scalable systems.

## Incident Response
Structured process for detecting, analyzing, and recovering from cybersecurity incidents.

## Evidence dates
Separate dates for documentation, version, pricing, data terms and an actual scoped trial. A null value means not established in this catalog, rather than false or safe.

## Cost per accepted task
All attributable costs over a defined period divided by accepted tasks in that same period. Human corrections, operations and model usage belong in the numerator.

## MCP
Model Context Protocol connects clients to tools and resources and defines optional transport authorization. Implementations must enforce access rights; retrieved instructions remain untrusted. Source: https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization

## Agent permissions
Explicit limits on what an agent can read, write, execute, send or spend. These boundaries require enforcement by the connected system, not only a prompt or status field.