A new OECD Due Diligence Guidance for Responsible AI falls short of expectations, but still provides a useful resource.
During the 2023 revision of the OECD Guidelines, OECD Watch members fought hard for new expectations for companies to address the harmful impacts of technology across their value chains.
For two years after the revision, we worked with civil society and union partners to advocate the strongest possible text in new AI due diligence guidance. Read more about civil society’s advocacy here.
We are disappointed the final text fails to identify the myriad human and environmental risks from AI – a serious gap in the context of widening AI reliance. However, the guidance still contributes by reframing AI due diligence through the lens of the OECD’s risk-based approach.
Download the OECD’s AI Due Diligence Guidance here: https://www.oecd.org/en/publications/oecd-due-diligence-guidance-for-responsible-ai_41671712-en.html
Key features of the AI due diligence guidance include that it:
- Applies the OECD’s six steps of DD to AI, covering aspects missed in many parallel laws and standards, and outlining practical actions to implement each step for AI harms.
- Targets three broad “groups” of companies in the “AI life-cycle” – suppliers of AI inputs (financing, hardware, digital infrastructure, data, code, algorithms); companies designing, developing, and deploying AI; and AI users (from sellers to banks, manufacturers, hospitals, universities).
- Emphasises the importance of meaningful stakeholder engagement with impacted people during all due diligence steps.
Key gaps are that it:
- Fails to list and provide elaborative examples on the wide range of human rights, worker rights, environmental, anti-competitive, and other societal harms of AI.
- Fails to call out the nexus with government as AI users – and potential “mis-users” in repressive regimes.
- Implies, but fails explicitly to spell-out, the gaps in parallel AI standards and laws – particularly on stakeholder engagement and remedy – to urge recentring around the OECD’s six-step approach.
Using the guidance
Valuably, the OECD will launch online use-case models to apply the guidance to specific risks and user groups. OECD Watch will urge the OECD to target leading risks and company types.
OECD Watch recommends civil society use the guidance and future case models to underscore the OECD’s support for your advocacy for victims’ rights, stakeholder engagement, and remedy in both complaints and policy advising on AI harms.
