One word, several authorities.

A machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.

OECD AI Principles

An engineered or machine-based system that can generate outputs such as predictions, recommendations, or decisions, for a given set of objectives.

AI RMF 1.0, Glossary

Information processing technologies that integrate models and algorithms which produce a capacity to learn and to perform cognitive tasks.

UNESCO Recommendation, Definitions

Systematic error introduced into sampling, training, or use that results in unfair outcomes; categorized as systemic, computational, and human-cognitive in NIST SP 1270.

NIST SP 1270; AI RMF

Algorithmic bias as evaluated for unjustified disparate outcomes against affected groups, with mitigation evidence required.

CertifAIEd

Discusses bias under fairness and non-discrimination obligations rooted in international human rights law.

UNESCO Recommendation

Human oversight and determination — responsibility traceable to natural or legal persons across the lifecycle.

UNESCO §III.II

Operationalized through GOVERN function: roles, accountability, processes.

AI RMF GOVERN

Oversight is evidenced through Accountability roles and recourse mechanisms.

CertifAIEd Accountability

Composite measure of an event's probability of occurring and the magnitude of consequences. AI RMF emphasizes negative and positive impacts.

AI RMF 1.0, Glossary

OECD does not prescribe a risk definition; classification dimensions support whatever risk framework is applied.

OECD Classification

Risk in ATLAS is implicit: characterized via concrete adversarial behaviors and mitigations rather than a generic definition.

MITRE ATLAS

Adverse impacts on people (civil liberties, physical/psychological safety, economic opportunity), organizations, and ecosystems.

AI RMF 1.0, §3

Harms are framed against human rights and ecosystem flourishing; the principle of proportionality and do-no-harm applies.

UNESCO §III.II

First Nations data: information about First Nations and their lands, resources, and communities, governed by First Nations authority.

FNIGC OCAP®

Data and input characterized along provenance, dynamic-vs-static, structure, and rights status (Dimension 3).

OECD Classification

AI data: training, validation, test, and inference data; includes synthetic data and metadata about data.

AI RMF Glossary

Data and information about First Nations and their lands, resources, and communities, governed by First Nations authority. OCAP® principles apply.

FNIGC

Transparency and explainability — preconditions for protecting human rights; context-specific in form.

UNESCO §III.II

Stakeholder-tailored disclosure of capabilities, limitations, data, and decision logic.

CertifAIEd Transparency

Transparency & accountability are trustworthiness characteristics measured under MEASURE 2.

AI RMF MEASURE 2

Defined roles, responsibilities, and recourse for outcomes produced by the system.

CertifAIEd Accountability

Operationalized through GOVERN 1 and GOVERN 2: policies, processes, and accountability structures.

AI RMF GOVERN

Responsibility and accountability traceable to physical persons or legal entities across the lifecycle.

UNESCO §III.II

An entity with capability and intent to cause harm to ML systems; characterized by tactics, techniques, and procedures.

ATLAS

Users, affected stakeholders, and human rights / democratic values are characterized in Dimension 1.

OECD Classification, Dim 1

AI actors: those who play an active role in the AI lifecycle, including affected communities.

AI RMF, Actors