AI, Digital Finance, and the Next Financial Stability Test
AI and the digitalisation of finance are moving faster than the guardrails around them. A senior executive at the Bank for International Settlements warns they are making it harder for central banks to spot, price, and contain system-wide risks.
The headline issues are clear: market functioning under stress, operational resilience across shared infrastructure, and the speed at which shocks can spread. For finance leaders, this isn't theoretical. It's now a risk-management problem with real liquidity, counterparty, and compliance consequences.
Where the pressure builds
Speed amplifies shocks. AI-driven strategies and tokenised markets compress time. Execution accelerates, liquidity thins at the edges, and price moves can escalate quickly. When models chase the same signals, exits get crowded and volatility spikes.
Concentration in the tech stack. Modern AI depends on a small set of providers: specialised chips, cloud compute, external data, and pretrained models. Tokenisation sits on shared platforms, protocols, custodians, and oracles. Outages, cyber incidents, or vendor failures in any one layer can ripple across portfolios and venues.
New interdependencies. Tokenised assets connect brokers, custodians, settlement layers, and compliance services in unfamiliar ways. These links often run across borders, making fragmented rules and inconsistent oversight a real weakness. As the BIS message puts it: "Because these developments cut across jurisdictions, fragmented or inconsistent approaches to governance can create gaps and frictions."
Practical steps for banks, asset managers, and market infra
- Map critical dependencies: Build and maintain an inventory covering cloud regions, model providers, data feeds, token platforms, oracles, and middleware. Identify single points of failure and substitution paths.
- Diversify and plan exits: Use multi-cloud or multi-region where feasible, negotiate portability in contracts, and run periodic exit drills to validate you can switch providers under stress.
- Control AI execution risk: Set guardrails for model-driven trading: throttle parameters, pre-trade checks, market-imbalance triggers, and hard kill switches. Monitor liquidity across venues in real time; test thresholds in live-fire exercises.
- Strengthen model risk management: Validate models against regime shifts, track data lineage, detect drift, and run scenarios for correlated behaviour and feedback loops. Document override policies.
- Boost liquidity resilience: Pre-arrange intraday liquidity buffers and collateral mobility. Define playbooks for tokenised collateral and settlement delays on shared ledgers.
- Harden operations and cyber: Conduct joint failover tests with critical vendors, rotate credentials and restrict privileged access, and test incident response across token platforms and APIs.
- Clarify legal risk: Define liability across smart contracts, token bridges, and service providers. Ensure settlement finality and dispute resolution are documented.
- Coordinate cross-border: Agree incident-reporting protocols with counterparties, CCPs, and custodians. Align data-sharing and escalation paths ahead of time.
- Elevate governance: Put AI and tokenisation risk on the board agenda. Set risk appetite, assign accountable owners, and ensure independent oversight of models and third parties.
What supervisors may focus on next
- Expectations for third-party concentration risk and "critical service provider" regimes.
- Common taxonomies and data-sharing for operational incidents and cyber events.
- Principles for AI model risk in finance, including transparency and testing standards.
- Interoperability and settlement finality standards for tokenised assets and platforms.
- System-wide stress testing that accounts for model herding and token market liquidity.
- Closer cross-border coordination to avoid gaps and conflicting rules.
Why this matters now
AI and tokenisation can make markets faster and more efficient, but they also tighten the coupling between firms and infrastructure. That's great when conditions are calm-and unforgiving when they're not.
The takeaway for finance teams: treat the tech stack like core market infrastructure, not a black box. Build optionality before you need it, test it under stress, and assume your counterparties are using similar models that may react the same way at the worst possible time.
Further reading
Upskilling your team
If you're formalising AI literacy across risk, markets, and ops, see our curated list of finance-focused tools and courses: AI tools for finance.
Your membership also unlocks:
