New OECD guidance for organizations to shape their AI governance framework

AI is now a corporate reality. Organizations have experimented with various governance models, but they need to evolve beyond basic compliance to navigate complex global regulations and ethical considerations around fairness and transparency. To succeed, organizations require a strategic framework that embeds oversight, technical bias audits, and cultural training throughout the entire AI lifecycle. 

The newly released OECD due diligence guidance provides a roadmap for achieving this. Organizations can structure their governance implementation plan around the following framework: 

1. Policy framework and management systems: Establish foundational policies that reflect core principles: human-centered AI, fairness and non-discrimination, transparency and explainability, robustness, security, safety, and accountability. Build supporting governance structures and management systems to operationalize these principles.

2. Risk identification and assessment: Conduct comprehensive risk scoping and assessment, supported by meaningful stakeholder engagement.

3. Risk prevention and mitigation: Implement responsible data practices, ensure transparency and explainability, and maintain security, robustness, and responsible deployment standards.

4. Tracking and monitoring: Establish processes for ongoing tracking, testing, and evaluation, along with thorough incident documentation.

5. External and internal communication: Develop audience-appropriate disclosures and ensure compliance with regulatory reporting requirements.

6. Remediation planning and mechanisms: Create clear pathways for addressing issues and providing remedies when harms occur.

As AI increasingly touches multiple functions and departments within an organization, a siloed approach to governance is no longer viable. Rather than implementing a standalone AI governance framework, organizations should integrate AI governance into their existing compliance and risk management structures.