Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsSome links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Pension funds assess AI investment risk by looking across the whole portfolio—not just at companies labelled “AI”—and testing how an AI-related market shock could affect returns, liquidity, funding and members. Trustees also examine how investment managers and service providers use AI, including their controls for model errors, cyber incidents and reliance on a small number of technology vendors.
The UK Pensions Regulator (TPR) advises trustees to identify direct and indirect AI exposure, test a severe but plausible downturn in AI-related assets, and compare potential losses, liquidity demands and changing correlations with existing stress tests. That approach applies to AI’s possible effects on investments and markets as well as to AI tools used in financial services.
As an Amazon Associate I earn from qualifying purchases.
Why does AI investment risk require a portfolio-wide assessment?
AI-related risk is not confined to a fund’s holdings in technology companies. It can also reach a pension portfolio through semiconductor manufacturers, cloud providers, utilities, energy assets, real estate, data-centre financing, private markets, credit and companies that depend on AI infrastructure. A holding may therefore be exposed even if its fund or company is not marketed as an AI investment.
Exposure can accumulate in ways that are easy to miss. A market-cap-weighted index can allocate more to AI-linked firms as their market values rise. Separately managed active funds may own many of the same companies. Investments in infrastructure, lenders or suppliers can add exposure through a different route. Looking at each fund in isolation can make a portfolio appear diversified while leaving it sensitive to the same companies, financing conditions or infrastructure bottlenecks.
#1 Best Overall
TPR’s guidance calls for trustees to look through the scheme’s whole portfolio, including default arrangements. Its 22 October 2026 press release reported that global private AI investment rose 127.5% year on year in 2025. That figure describes private investment globally; it is not a measure of pension-fund exposure or a forecast of future returns. The growth is one reason to make sure private assets and debt are included in the review, rather than relying solely on public-market labels.
AI investment risk is also broader than the chance that a company’s share price falls. A market repricing can coincide with tighter liquidity, increased collateral calls, financing pressure or a rise in correlations between assets that were expected to diversify one another. For a pension scheme, the relevant question is how those effects interact with its assets, liabilities, cash needs and member outcomes.
What exposures should trustees look for?
Trustees need an exposure map that distinguishes direct holdings from indirect dependencies and shows where several routes lead to the same risk. The map should cover the full scheme, including default options, private assets and debt—not only funds whose names mention AI.
Direct and indirect links
- Direct exposure: shares, debt or private investments in businesses whose revenues or valuations are closely tied to AI development or adoption.
- Supply-chain and infrastructure exposure: semiconductors, cloud services, data centres, utilities and energy assets that support AI investment or operations.
- Financing exposure: lenders, private credit and other investments that may be affected by the funding needs or financial condition of AI-related businesses and infrastructure.
- Portfolio overlap: repeated holdings across mandates, index funds and active strategies, plus common financing or service-provider dependencies.
These categories are prompts for look-through, not a claim that every company in them has the same exposure. Trustees should ask managers to explain how they define AI-related exposure and what evidence supports that classification. A company can have meaningful dependencies without reporting AI as a separate business line; conversely, a broad label may not reveal how much its prospects depend on AI.
Concentration and diversification
Assess whether rising market values have increased the portfolio’s weight in a small group of AI-linked companies, and whether different managers hold the same names. Then consider concentration beyond securities: multiple investments may depend on a common cloud provider, chip supply, energy source or data-centre market. Such shared dependencies may matter most in a disruption, when diversification assumptions are under strain.
How should pension funds stress-test AI investments?
A useful stress test asks what could happen under a severe but plausible AI-related shock, then measures the consequences for the scheme rather than reporting a market-loss estimate alone. TPR specifically advises trustees to compare losses, liquidity demands and correlation changes with their existing stress tests.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Build scenarios around plausible failure paths
Scenarios should reflect the scheme’s actual exposures. Relevant possibilities include an AI-related repricing, a reversal or delay in infrastructure investment, slower-than-expected adoption, a regulatory shock, a supply bottleneck, energy constraints or capital spending that exceeds what businesses can sustain. These are scenario inputs, not predictions that any one event will occur.
For each scenario, make assumptions explicit: which holdings or revenue streams are affected, how quickly the shock unfolds, whether financing remains available and whether other assets move at the same time. Test a base case alongside adverse and severe-but-plausible cases so trustees can see how outcomes change when assumptions worsen. The point is not to produce a single precise forecast; it is to expose vulnerabilities and show which assumptions drive the result.
Measure more than asset-price losses
Ask managers and advisers to report the effects in terms relevant to the scheme. Depending on its structure and investments, that can include:
- Estimated losses by asset, mandate and the portfolio as a whole.
- Liquidity needs, including whether assets can be sold or cash raised when required.
- Leverage, financing pressure and potential collateral demands.
- Funding implications for a defined-benefit scheme and the effect on benefit security.
- Changes in correlations that could weaken expected diversification.
- For defined-contribution arrangements, the possible effect on members’ account values and decisions to transfer, switch or withdraw.
Results should be compared with existing stress tests, as TPR recommends. That helps a committee understand whether an AI-linked shock creates a genuinely different vulnerability, intensifies one it already tracks or falls within the scheme’s established risk limits. Trustees should record the assumptions and decision thresholds used, not just the headline loss.
How do trustees assess valuations and market assumptions?
Valuation review means asking whether expectations about earnings, productivity and adoption justify the prices and investment plans on which portfolio exposures depend. It does not require trustees to decide which AI businesses will succeed; it does require them to understand what assumptions managers are relying on and how sensitive their conclusions are to those assumptions.
Challenge the pace and scale of expected adoption, the productivity gains assumed, the ability to convert investment into earnings, and the capital spending required to build or operate infrastructure. Ask what happens to valuations if adoption is delayed, energy is constrained, a supply bottleneck persists, regulation changes or investment overshoots demand. Where a manager’s view depends on a particular outcome, request the evidence and a clear account of what would change that view.
The Financial Stability Board (FSB), in its 14 November 2024 assessment, identified possible changes to market structure, macroeconomic conditions and energy use alongside risks from correlations, third-party dependencies, cyber incidents, model problems, data quality, fraud and disinformation. This makes market structure and common dependencies relevant parts of scenario analysis—not just the prospects of individual listed companies.
What should trustees request from investment managers?
Trustees can use due diligence to make exposures and risk controls comparable across managers. A manager’s answer should show what it assessed, what it could not assess and what action follows if an assumption or control fails.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Portfolio and scenario information
- Look-through holdings across public and private assets, including debt where relevant.
- The manager’s definition of direct and indirect AI exposure, the classification method and its limitations.
- Repeated holdings, index weights, leverage and financing links, plus dependencies on cloud, chip, energy or data-centre providers.
- Scenario assumptions, estimated losses, liquidity effects, collateral or leverage impacts and correlation changes.
- How valuation, adoption, productivity, energy and regulation assumptions affect the manager’s conclusions.
Controls for AI systems and service providers
Where a manager or its providers use AI, ask how models are validated and monitored, where data comes from, how data quality is checked, how decisions can be explained and whether a human can intervene. Ask about bias, privacy, cyber security, third-party dependencies, incident reporting, business continuity and plans for changing or exiting a provider. These questions are especially important where a failure could affect multiple funds or functions at once.
FINMA’s guidance identifies model robustness, correctness, explainability and bias, as well as data security, quality and availability, IT and cyber risk, third-party risk, and legal and reputational risk. These categories offer a practical due-diligence checklist; their applicability will depend on the manager, use case and relevant jurisdiction.
What risks arise when a manager uses AI?
There are two connected but distinct questions: whether an investment is exposed to AI-related market outcomes, and whether AI tools used to make or support investment decisions are reliable and appropriately governed. A scheme can have either exposure without the other, and trustees should not treat one review as a substitute for the other.
AI-supported analysis or investment services can create risks through opaque outputs, bias, overreliance on automated recommendations, privacy or security failures and weak human oversight. ESMA says use of AI in investment services remains subject to the organisational, conduct and best-interest duties under MiFID II. That is a regulatory reminder for firms within the relevant framework, not a statement that every pension scheme or provider is governed identically.
Rank #4
The US Treasury recommends reviewing each AI use case for compliance with existing law before deployment and reassessing it periodically. A trustee’s due diligence should therefore ask what the tool is used for, who is accountable for its output, what testing and monitoring take place, how errors are escalated and whether the provider can demonstrate compliance with applicable requirements. An adviser or manager’s use of a tool does not transfer trustee accountability for scheme governance.
How can committees turn the review into an ongoing process?
A practical process assigns responsibility, gathers consistent information, tests decisions against the scheme’s risk appetite and specifies when issues must be escalated. The review should be repeated as exposures, managers, systems and market conditions change.
- Set the scope. Identify which portfolios, default arrangements, private assets, debt and service providers must be covered. Name the committee member or function responsible for coordinating the review.
- Build the exposure taxonomy. Define how the scheme will classify direct holdings, indirect links, financing, infrastructure dependencies and AI use by managers. Ask every manager for information using consistent definitions where possible.
- Reconcile common exposures. Compare look-through data across mandates. Identify repeated holdings, index concentration, financing links and shared dependencies on cloud, chips, energy or data centres.
- Run and review scenarios. Use base, adverse and severe-but-plausible cases. Report losses, liquidity, leverage, collateral, funding and correlation effects, and explain which assumptions produce the largest changes.
- Challenge the response. Compare the results with existing tests and risk appetite. Decide whether to seek better information, change monitoring, discuss a mandate with a manager or take another action under the scheme’s governance process.
- Check operational controls. Review model validation, data provenance and quality, explainability, bias, human oversight, cyber security, privacy, incident response and continuity arrangements for relevant providers.
- Set triggers and review dates. Document who receives an escalation, what kind of change prompts it and how often the committee revisits exposures and controls. The appropriate frequency depends on the scheme’s circumstances and governance arrangements.
TPR’s guidance places responsibility on trustees to maintain investment governance and manage risk over the short, medium and long term. Advisers and managers can supply analysis and carry out delegated work, but trustees should retain oversight, challenge the assumptions and document decisions. TPR’s 22 October 2026 press release quoted Executive Director of Market Oversight Ben Gunnee saying it expects schemes to maintain high investment-governance standards and manage risk over those time horizons.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How should schemes account for members and operational disruption?
Investment stress is only one part of resilience. Cyber incidents, data leakage, model drift, vendor outages and fraud can disrupt operations or expose member information. AI-generated scams and disinformation can also complicate communications during a volatile period. Managers and providers should be able to explain their incident response and continuity arrangements, including how the scheme would be notified and how essential services would continue.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →For defined-contribution schemes, a market downturn can coincide with members making consequential decisions, such as crystallising losses or stopping contributions. Trustees should have clear, timely communication procedures that explain what has happened, what members can do and where they can obtain reliable help. Communications should not imply that short-term market movements can be predicted or promise outcomes that the scheme cannot control.
How should trustees compare managers and portfolios?
There is no single score that captures AI risk. A side-by-side review is more useful when it separates exposure, market sensitivity and the quality of information and controls. The table below is a committee framework; it does not assign ratings or imply that a particular asset class or strategy is inherently safer.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
| Comparison area | What to establish | Why it matters |
|---|---|---|
| Exposure | Direct and indirect links; public and private holdings; debt and infrastructure dependencies | Reveals risks that may not appear in a fund name or top-holdings list |
| Concentration | Repeated holdings, passive index weights and common dependencies across managers | Shows whether several positions could be affected by the same shock |
| Valuation sensitivity | Dependence on adoption, earnings, productivity, energy and regulatory assumptions | Identifies what must go right for current expectations to hold |
| Liquidity and leverage | Ability to raise cash, financing arrangements and potential collateral needs under stress | Connects market declines to the scheme’s ability to meet obligations |
| Stress behaviour | Losses and correlations under comparable scenarios | Shows whether diversification may weaken when it is most needed |
| Provider and model controls | Third-party reliance, data and model validation, human oversight and incident plans | Tests operational resilience as well as investment analysis |
| Governance and outcomes | Regulatory compliance, stewardship, escalation arrangements and member effects | Connects technical risk assessment with fiduciary oversight and scheme decisions |
Ask managers to distinguish missing data from a genuine absence of exposure. If look-through is incomplete, record which assets or dependencies are not visible, how that limits the assessment and whether the manager can improve reporting. A confident-looking total is not useful if managers have used incompatible definitions.
Frequently Asked Questions
Does a pension fund need to own AI stocks to be exposed to AI risk?
No. Exposure can arise through suppliers, infrastructure, lenders, private assets or shared technology providers. The relevant test is whether the investment’s value, financing or operations could be affected by AI-related developments, not whether its name includes AI.
Recommended Free Tools
Are pension trustees responsible for an adviser’s AI tool?
Delegating work does not remove trustees’ accountability for scheme governance. Trustees should understand what an adviser’s tool is used for, what assurance supports it and who reviews or challenges its outputs.
Does a severe AI stress scenario mean trustees expect an AI crash?
No. Stress testing examines whether a scheme could withstand a plausible adverse event; it is not a forecast that the event will happen. Its value lies in revealing vulnerabilities and decisions that may need attention before conditions worsen.
Can one AI risk score compare every pension fund?
A single score would obscure differences in holdings, liabilities, liquidity needs and provider arrangements. Trustees should compare consistent evidence and scenarios while interpreting the results in the context of each scheme.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
