The Governance of AI: Examining Systems Stewardship in Practice
Cora Buentjen, Josie Zenger and Rory Sullivan
On 11 September, Chronos Sustainability and Railpen hosted a webinar to launch Navigating AI: An investor framework for assessing exposure and governance. The session examined how companies are managing and overseeing AI deployment and development, exploring how investors can assess AI-related risks and opportunities. It also used AI to explore the question of what systems stewardship might look like in practice.
Reflection 1: Universal Ownership Guides Decision-making
Railpen's starting point is that it is a universal owner. That is, as a consequence of its being a large, diversified, long-term owner, it effectively owns a representative slice of the entire global economy. Therefore, its investments are exposed to, and cannot avoid, systemic risks (i.e. those risks that transcend sectors and jurisdictions, such as climate change, economic crises, and governance failures, and can therefore not be diversified away).
Universal ownership points investors to two related responsibilities. The first is to ensure that individual companies are managing these risks effectively. The second is to support markets and institutions in responding to them in ways that sustain long-term value creation.
The Railpen team shared five practical tips for putting this into practice:
Follow the evidence, so that priorities rest on analysis rather than assumptions.
Prioritise by theme, rather than attempting to address every systemic issue at once.
Focus on companies that shape the system, not only the laggards or the leaders.
Use policy and market signals where company-level engagement alone is insufficient.
Draw on operational expertise from across the organisation.
Reflection 2: AI is a Systemic Issue and Should be Managed as Such
AI is rapidly transforming how companies in all sectors operate. It creates significant opportunities but also creates risks that extend well beyond individual firms and that cannot be diversified away.
The webinar focused on three key examples:
Digital concentration. AI ecosystems (the interconnected network of data, hardware, software models and developer tools that build and govern AI) are increasingly concentrated among a small number of providers, creating scale advantages but also single points of failure and dependency risks.
Cascading operational disruption. AI can improve efficiency within critical services, yet model failures or correlated errors can spread quickly. For instance, an error in a widely used model could affect customer services, logistics systems and decision-making tools across many businesses at once, causing widespread disruptions.
Reliance on shared infrastructure. AI adoption is driving demand for digital and energy networks, raising questions about capacity, access and resilience. The data centres that power AI need large amounts of electricity to run and, often, water to cool them. Where many facilities cluster in the same region, they can put pressure on local electricity grids and water supplies, competing with households, agriculture and other industries, thereby pushing up costs for other users.
These interdependencies create risks that a universal owner, such as Railpen, cannot diversify away. Addressing them requires action beyond any single portfolio. At the company level, investors need to understand how AI-related risks are being assessed and managed. At the market level, they need to support consistent standards, better data and policy frameworks that keep pace with the technology.
Reflection 3: A Systemic Focus Does Not Mean Ignoring Company-specific Risks
As part of our work, we assessed how these systemic risks can be understood at a company-level.
To this end, we developed an AI Governance Framework (AIGF), which was structured around three pillars: governance and strategy, risk management, and performance reporting. Applying it to Railpen's actively managed portfolio – a universe of 33 companies across a range of sectors including IT, financials and retail - showed that AI governance maturity is uneven, and often underdeveloped relative to risk exposure. The most telling distinction was between stated intent and evidence of implementation. Many AI policies signalled high-level alignment with responsible AI principles; however few companies had – at the time of our research - moved beyond this towards detailed policies, transparent disclosure, incident reporting and meaningful stakeholder engagement. This may simply be a reflection of the relative novelty of AI as a governance issue, or it may be symptomatic of weaknesses in how certain companies are approaching the governance of AI.
Conducting an assessment is, of course, only part of the picture. Investors then need to decide where engagement would be most effective and appropriate. One of the key reflections from our work is that – especially given that it is a nascent and fast-moving issue - investors engaging on AI must recognise that companies are at different stages of their AI governance journey, and that their exposure to AI-related risks and opportunities varies considerably. For investors with stakes in many companies, we suggested that companies with high exposure to AI risks combined with weak governance should be prioritised. This allows investors to concentrate their resources where the potential system-wide effects are greatest, without placing an unnecessary burden on lower-exposure companies.
Our Reflections: Three Wider Lessons for Systems Stewardship
What began as a question about AI governance produced three lessons that apply to systems stewardship more broadly:
Systemic stewardship does not mean every tool, every time. An issue can be systemic without requiring an investor to deploy every stewardship tool available, or to take the lead. Decisions should reflect capacity, resources and the value that leadership would add.
Approaches must keep pace with the evidence. AI adoption accelerated markedly in the year between our initial research report (Achieving effective AI governance – from theory to practice) and the benchmarking exercise presented in Navigating AI: an investor framework for assessing exposure and governance. Systemic issues evolve, and research and priorities need to be revisited accordingly.
Addressing systemic risks requires effective governance. Systemic issues rarely unfold as expected, and no investor can predict exactly how the risks will materialise. What investors can assess is whether companies have the skills, oversight and processes to remain resilient as circumstances change.
Related resources
AI Governance Framework: Systems Stewardship in Practice Webinar
Navigating AI: an investor framework for assessing exposure and governance
Photo by Igor Omilaev
