⚡ Quick Summary
Published by the World Economic Forum in collaboration with CPP Investments Insights Institute, this June 2024 white paper sets out how large investors can accelerate the adoption of responsible AI (RAI) across their portfolios, their investment partners and the wider ecosystem. RAI is defined as the practice of designing, building, deploying, operationalizing and monitoring AI systems in a manner that empowers people and businesses and impacts customers and society equitably, covering ethical and technical principles such as validity and reliability, safety, fairness, security and resilience, accountability and transparency, explainability and interpretability, and privacy.
The playbook argues that RAI both reduces risk and promotes growth, citing McKinsey & Company's nine categories of AI risk, the NIST AI Risk Management Framework, the EU AI Act's risk-based approach and its fines of up to €35 million or 7% of worldwide annual turnover, and findings that companies with a comprehensive responsible approach earn twice as much profit from their AI efforts and nearly 30% fewer AI failures. Its practical core is a three-step engagement process, illustrated by case studies of Radical Ventures, Credo AI, NBIM, UBS Asset Management, Manulife and Temasek and by sample engagement objectives, tools and examples. It closes with the hurdles ahead, including standardised metrics, capacity building and regulatory clarity.
🧩 What’s Covered
- Introduction: frames RAI as value-preserving and value-creating for investors, traces AI's history from the 1956 Dartmouth project, cites a 2023 McKinsey estimate of up to $25.6 trillion in annual economic impact, and in Box 1 defines RAI and lists its seven principles, noting they are context dependent and involve trade-offs.
- Business case (Section 1): Table 1 groups AI risk into ethical and social, technical and operational, and security and legal categories, with sub-risks including privacy violations, bias and discrimination, inaccurate output, third-party risk, IP infringement and security threats; Figure 1 (NIST AI RMF) maps harm to people, organizations and ecosystems; examples include Zillow's approximately $881 million loss in 2021 and the New York Times lawsuit against OpenAI and Microsoft.
- Regulation and value: at least 148 AI-related bills passed since 2016, the EU AI Act described as the most stringent AI regulation, its extraterritorial scope and "Brussels effect", and value claims such as twice the profit from AI efforts and 30% fewer AI failures.
- Stakeholders (Section 1.2, Table 2): maps governments and regulators, professional and research organizations, asset owners, asset managers, company boards and management to their role in RAI and their incentive to accelerate it.
- Engagement steps (Section 2): Step 1, develop RAI commitments and apply them to internal operations; Step 2, conduct RAI due diligence on the portfolio; Step 3, engage with companies, external managers and the broader ecosystem, with AI governance identified as the key point of leverage.
- Engagement with companies (Section 2.2, Table 3, case studies 3–4): covers AI principles, policies and procedures, roles and responsibilities, and investor transparency, with NBIM's three elements of RAI and UBS Asset Management's three-layer AI oversight model.
- External asset managers (Section 2.3, Box 3, Table 5, case study 5): supplies a discussion guide of questions for asset owners and describes Manulife IM's responsible technology innovation work and its general partner questionnaire.
- Ecosystem and hurdles (Section 2.4, Table 6, case study 6, Section 3): Temasek's AI Pod and AI Verify Foundation participation, plus gaps: dynamic governance frameworks, standardised RAI metrics, capacity building, financial materiality of RAI, stakeholder alignment, regulatory clarity, ESG labels and the tension between RAI and corporate imperatives.
💡 Why it matters?
For asset owners, asset managers, board directors and stewardship teams, the playbook turns a broad principle into a sequence of engagement actions: set internal RAI commitments, run due diligence to locate where AI drives core revenue or sits in high-risk and regulated areas, then press companies on AI principles, policies, board competence, transparency and metrics. Its tables supply ready-made objectives, supporting tools and examples, including the NIST AI RMF, ISO/IEC 42001, the OECD AI Principles, the ICGN engagement guide and the ICGN-GISD Model Mandate. It also shows how emerging regulation such as the EU AI Act shifts the incentive to act before rules settle.
❓ What’s Missing
The playbook is deliberately non-exhaustive: it states that it is not a comprehensive view of all levers available and does not re-examine common engagement tactics such as shareholder resolutions, proxy voting and negotiating terms into investment management agreements, referring readers to UN PRI guidance instead. It offers no standardised RAI metric, noting that authoritative sources such as MSCI and the ISSB have yet to address AI specifically and that evaluating process metrics is a stopgap. Practical detail stays generic: the tables give sample objectives and examples rather than thresholds, and several hurdles — regulatory certainty, financial materiality, the politics of ESG labels and short-termism — are left open.
👥 Best For
Best for asset owners, stewardship and ESG engagement teams, and investment managers building an RAI engagement programme, as well as board directors of portfolio companies who need to know what investors will ask about AI oversight. It also suits sustainable-investing and policy analysts who want a compact map of the investor-facing debates around AI governance.
📄 Source Details
Responsible AI Playbook for Investors (white paper), published by the World Economic Forum in collaboration with CPP Investments Insights Institute, June 2024; 25 numbered pages plus cover, 26 PDF pages. Lead authors: Chris Gillam (CPP Investments) and Devendra Jain (World Economic Forum); the document also lists community members, World Economic Forum and CPP Investments staff, and production credits. Language: English. No reference number or document URL is printed; the endnotes contain only links to cited external sources. The extracted text covered all 26 pages.
