Data, AI & Green IT glossary.

The EU Artificial Intelligence Act (Regulation 2024/1689), the world's first horizontal regulation of AI. It sorts systems into four risk levels — unacceptable, high, limited, minimal — with growing obligations. Full application of the high-risk rules on 2 August 2026.

The General Data Protection Regulation (EU 2016/679). Governs all processing of personal data in the Union: legal bases, minimisation, data-subject rights, impact assessments. It applies in full to processing carried out by AI systems.

A design approach seeking to deliver an AI system's business value with the minimum of compute, energy and data: right-sizing, model reuse, inference optimisation and systematic footprint measurement.

An approach aimed at reducing the environmental footprint of digital technology by mobilising only strictly necessary resources. It covers eco-design, architecture frugality and measuring consumption (kWh, tCO₂).

A generative-AI architecture that enriches a language model's answers with information retrieved from an organisation's own document base. It reduces hallucinations, allows sources to be cited, and keeps data under control.

A set of practices for industrialising Machine Learning: training, deployment, monitoring (drift) and retraining of models in a reproducible, automated way. It is what moves a model from notebook to production.

Management of reference data: creating a single, reliable, shared source for an organisation's key entities (customers, products, suppliers) to eliminate divergences between systems.

A hosting infrastructure whose data and operations remain under European jurisdiction and control, shielded from extraterritorial laws such as the CLOUD Act. It aims at strategic autonomy over sensitive data.

An impact assessment required under the GDPR when processing is likely to result in a high risk to individuals — common with AI. It describes the processing, evaluates the risks and sets the mitigating measures.

The ability of people who design, deploy or use an AI system to understand how it works, its capabilities, limits and risks. The AI Act (Article 4) makes it an obligation from February 2025.

A cloud financial-management discipline: making infrastructure costs visible and controllable through right-sizing, governance (tagging, budgets) and continuous trade-offs between performance and spend.

Applying FinOps principles to environmental footprint: measuring kWh and tCO₂ per service, choosing low-carbon-intensity regions and scheduling workloads at the most sober hours (carbon-aware).

Traceability of a data point's journey, from its source to its final use, through successive transformations. Essential for quality, audit and regulatory compliance.

Re-training an existing model on specific data to adapt its style or tasks. Distinct from RAG, which retrieves information at question time without changing the model; the two are often combined.

A language-model answer that looks plausible but is factually wrong or invented. RAG and source citation strongly reduce it; no approach removes it entirely, hence the need for guardrails and measurement.

The EU cybersecurity directive (EU 2022/2555) strengthening security and incident-notification obligations for a wide range of essential and important entities, including the public sector.

Designing digital services to reduce their footprint across the whole life cycle. It draws on references such as RGESN and GR491. An eco-designed service is often faster and cheaper too.

The roles, rules and processes ensuring an organisation's data is reliable, secure, compliant and usable: data owners, stewards, catalogues, quality policies and data contracts.