Adopted standard

OECD AI Principles

An Ethotechnics critique of the OECD AI Principles: non-binding by design, they ask AI actors to be accountable and set no duty anyone can enforce.

Ethotechnics critique

Where the principles fall short

  • Non-binding by design: Nothing in the principles can be tested, so a program can claim after it has caused harm that it followed them, and no one can show it did not.
  • No allocation of power: The principles rarely say who must answer a challenge, who can halt a system, or on what clock.
  • Intervention without mechanism: Principle 1.4(b) asks for mechanisms, “as appropriate,” so a harmful system can be overridden, repaired, or decommissioned “as needed.” It does not require a tested halt path, a restoration deadline, or a limit on how long a case can stay pending.

Failure mode

Legible virtue + uncorrected errors

Good narratives and training, and no one bound to act on what goes wrong.

The dominant failure mode is a system that sounds responsible on paper while the evidence of its errors reaches no one who has to change it.

Principles to clauses

Where an Ethotechnics clause makes a principle testable

Each row names the principle, the clause that bears on it, and what the clause requires that the principle leaves open.

Principle What it asks Clause What the clause adds
1.3(iv) Provide information that enables those adversely affected by an AI system to challenge its output. STD-02 §2.2 , STD-07 §4.2 The appeal path names the authority that holds reversal power, and an objection gets a revision or a reasoned refusal within a stated clock.
1.4(b) Mechanisms, as appropriate, so a system that risks undue harm can be overridden, repaired, or decommissioned safely as needed. STD-06 §1.3 , STD-06 §5.6 One owner who can halt the deployment, and a withdrawal path rehearsed on a published cadence.
1.5(b) Traceability of datasets, processes, and decisions, to enable analysis of outputs and responses to inquiry. STD-02 §5.3 , STD-07 §1.2 A decision log that links each challenged decision to its evidence, reviewers, and outcome, in records that are never edited.

Principle text from the OECD Recommendation on Artificial Intelligence, as revised in 2024, paraphrased.

Cite this page · APA, MLA, Chicago, BibTeX, RIS

Version

vDraft

Last updated

Oct 6, 2026

APA

Ethotechnics Institute. (2026). OECD AI Principles. Ethotechnics Institute. https://ethotechnics.org/standards/critiques/oecd-ai-principles

MLA

Ethotechnics Institute. "OECD AI Principles." Ethotechnics Institute, 2026, https://ethotechnics.org/standards/critiques/oecd-ai-principles.

Chicago

Ethotechnics Institute. "OECD AI Principles." Ethotechnics Institute. Oct 6, 2026. https://ethotechnics.org/standards/critiques/oecd-ai-principles.

BibTeX

@misc{ethotechnics_standards_critiques_oecd_ai_principles,
  title={OECD AI Principles},
  author={Ethotechnics Institute},
  year={2026},
  howpublished={Ethotechnics Institute},
  url={https://ethotechnics.org/standards/critiques/oecd-ai-principles},
  version={Draft}
}

RIS

TY  - WEB
TI  - OECD AI Principles
AU  - Ethotechnics Institute
PY  - 2026
UR  - https://ethotechnics.org/standards/critiques/oecd-ai-principles
ER  -