OECD
Classify AI systems across People & Planet, Economic Context, Data & Input, AI Model, Task & Output.
- Audience
- Policymakers, regulators, organizations.
- Unit of analysis
- AI system characteristics.
- Lifecycle coverage
- Cross-cutting characterization.
- Outputs
- Classification profile.
- Strengths
- Structured taxonomy; comparable across systems and jurisdictions.
- Cautions
- Descriptive, not prescriptive; does not tell you what to do about risk.
- Jurisdictional scope
- OECD member states + adherents (50+ countries).
- Evidentiary weight
- Descriptive; widely used in regulatory taxonomies (e.g. EU AI Act draws on it).
- Cost to adopt
- Low — typically a one-time profiling exercise per system.
- Certification path
- None; classification only.
First published February 2022. Builds on the OECD AI Principles (2019, updated 2024). Used as a common vocabulary for AI policy.
Framework for the Classification of AI Systems
Indexed at the structural level. Excerpts are quoted under fair-use; full text is linked, not rehosted.
Subcategories11
- Dimension 11.1 Usersframingdeployment
Users of the AI system
“Who deploys or interacts with the system; expertise; rights to opt out or contest.”
- Dimension 11.2 Affected stakeholdersframingmonitoring
Affected stakeholders
“Individuals, groups, communities, and ecosystems materially affected by system outputs, including non-users.”
- Dimension 11.3 Human rights & democratic valuesframingdeployment
Human rights and democratic values
“Potential impact on rights, civic participation, and democratic processes.”
- Dimension 22.1 Sectorframing
Industry sector
“The sector and any sector-specific regulation that bears on the system (e.g., finance, healthcare, public sector).”
- Dimension 22.2 Criticalityframing
Criticality of function
“Whether the system supports critical functions (safety-of-life, essential services).”
- Dimension 33.1 Provenancedata
Data provenance
“Origin and chain of custody for training, evaluation, and inference data.”
- Dimension 33.2 Dynamic vs. staticdatamonitoring
Dynamic or static input
“Whether inputs are real-time, drifting, or fixed datasets.”
- Dimension 44.1 Model typemodel
Model type and family
“Statistical, symbolic, neural, hybrid, foundation/general-purpose model.”
- Dimension 44.2 Explainabilitymodeldeployment
Explainability
“Degree to which the system's behavior can be explained to relevant stakeholders.”
- Dimension 55.1 Autonomydeployment
Autonomy of action
“Whether outputs are recommendations, decisions, or autonomous actions.”
- Dimension 55.2 Combining tasksdeployment
Combining tasks and actions
“Compositional behavior: agents, tool use, chained models.”
Dimensions05
- Dimension 1framingdeploymentmonitoring
People & Planet
“Considers stakeholders affected by the system, including workers, consumers, third parties, and the environment.”
- Dimension 2framing
Economic Context
“The economic and sectoral environment in which the AI system is deployed, including criticality and scale.”
- Dimension 3data
Data & Input
“Provenance, structure, scale, quality, and rights status of data and inputs used by the system.”
- Dimension 4model
AI Model
“Model characteristics including type, training, inference, performance, and explainability properties.”
- Dimension 5deployment
Task & Output
“What the system does, the action it takes or recommends, and the autonomy with which it does so.”