OECD Due Diligence Guidance for Responsible AI (2026)
Step 3: Cease, prevent and mitigate adverse impacts – OECD Due Diligence Guidance for Responsible AI (2026)

OECD Due Diligence Guidance for Responsible AI (2026) 3.1.G2-transp: 3.1.G2-transp Mitigation by developers: transparency, explainability and traceability

Enable transparency, explainability and traceability over data sourcing, datasets, processes and development decisions, human review and appeal (Box 2.8: document use and risk information, data sources and processing, code and reproducibility details, how outputs are used and disclosed, monitoring strategy, limitations and biases, tailored through model cards, data sheets and system cards); include input data type and source, transformation, decision criteria and an AI disclosure in explanations; give clear, accessible explanations of consequential automated decisions; deploy content authentication and provenance where feasible (Box 2.9); contribute to measurement science; publish a guide for external stakeholders; and make disclosures clear enough for deployers and users to interpret outputs, backed by robust documentation.

Maintained by Gerard Blokdyk

Other controls in Step 3: Cease, prevent and mitigate adverse impacts – OECD Due Diligence Guidance for Responsible AI (2026)

Query this from an agent

The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.