FSB Stablecoin Risks and AI Financial Safeguards
The FSB warns G20 policymakers about stablecoin oversight gaps and AI-driven financial risks. See what the proposals mean for markets and institutions.

Stablecoins and AI: The FSB's Two Financial Stability Tests
Stablecoins and artificial intelligence are often treated as separate topics. One is about crypto markets and payment rails. The other is about automation and productivity. But for financial regulators, the underlying question is the same: can a fast-moving technology create risks that existing oversight can't quite contain?
That's the core issue in a letter the Financial Stability Board (FSB) sent to G20 finance ministers and central bank governors ahead of their October 15 meeting. The FSB pointed to gaps in cross-border stablecoin oversight and warned that AI-enabled cyber threats could put pressure on critical financial institutions and market infrastructure. It also urged G20 members to back the FSB's finalized practices for responsible AI adoption in finance.
The risk is concrete. A stablecoin redemption shock can feed into banks and traditional markets through reserve sales and deposit withdrawals. AI-related operational failures can hit firms that rely on shared technology providers. This article breaks down what the FSB flagged, why these risks can spill across market lines, and what the recommendations could mean in practice for regulators, financial firms, and crypto businesses.
Why the FSB raised stablecoins and AI together
The FSB's letter puts two technology topics side by side: stablecoin regulation and the use of AI across financial services. They aren't the same risk, and they don't need identical controls. What they do share is speed and reach - both can operate across institutions and borders faster than fragmented rules and crisis procedures can respond.
The FSB is an international body focused on global financial stability. It isn't a "world regulator" that directly supervises every bank, exchange, or stablecoin issuer. Instead, its influence comes through standards, monitoring, and recommendations that national authorities can adopt through their own frameworks. So even when the FSB calls for stronger safeguards, that's not automatically the same thing as a new enforceable rule for every participant.
In the letter, FSB Chair Andrew Bailey said cross-border stablecoin oversight and crisis management are still immature - or, in some places, haven't been fully developed. The FSB's 2025 review found that only 11 out of 28 jurisdictions it surveyed had finalized stablecoin regulatory frameworks. That suggests uneven readiness, not a complete account of every country's rules or how well each one works.
This matters when a token, its issuer, its reserves, and its users sit in different places. One jurisdiction may regulate the issuer, another may hold reserve assets, and users may rely on exchanges or payment services that operate elsewhere. If something goes wrong, authorities need clarity on who can act, what information is available, and how decisions will be coordinated.
AI creates a similar problem. A financial institution might build a model internally, run it on a cloud platform supplied by a third party, and deliver the service to customers in multiple markets. Tracing responsibility can get messy when systems, data, vendors, and regulated entities cross jurisdictions.
So the point of the FSB's message isn't simply "don't use new technology." It's about making the risks easier to identify and manage. In both cases, the key test is whether supervisors and firms can spot exposures early - and coordinate effectively when routine controls stop working.
How stablecoin stress can reach traditional markets
A stablecoin is typically designed to hold a stable value against a reference asset, often a fiat currency. In practice, that stability depends on confidence in the issuer's redemption arrangements and, when applicable, the assets backing the token. If holders start to doubt they can redeem at the expected value, or on time, confidence can drop quickly.
The FSB's report points to a clear path from a stablecoin run to traditional finance. If an issuer needs to meet a surge in redemptions, it may have to sell reserve assets. When many holders try to exit at once, forced sales can add pressure to the markets where those assets trade. The size of the impact depends on reserve composition, issuer liquidity, and broader market conditions. The concern isn't that every redemption automatically destabilizes everything - it's that rapid, concentrated redemptions can turn a crypto-market shock into a liquidity problem elsewhere.
There's also a deposit channel. Stablecoin issuers or reserve managers may hold cash and related balances at banks. If withdrawals spike to meet redemption demand - especially when flows are concentrated or occur during wider market stress - those banks can feel it too. Regulators are likely to want to assess this transmission channel alongside the issuer's own reserves and redemption process.
The FSB also flagged an additional complication: stablecoin structures with multiple issuers. Redemption demand might concentrate on one issuer even if reserves are held across different jurisdictions. That can create a mismatch between where the redemption obligation sits and where the resources supporting it are located. In that moment, authorities may struggle with questions about access to reserves, the applicable legal framework, and coordination across borders.
The operational questions behind a credible framework
For regulators and issuers, a usable framework has to cover more than whether a stablecoin is allowed. It should explain how the arrangement works when stress hits. For example:
- Who is responsible for honoring redemptions, and under what conditions?
- Where are reserve assets held, and how quickly can they be converted into cash?
- Which authorities supervise the issuer, custodians, payment services, and reserve managers?
- What information can supervisors access in normal operations and during a crisis?
- How will authorities coordinate if the issuer or reserve arrangements span borders?
These aren't just technicalities. They determine whether a promise to token holders can actually be carried out when markets are under pressure. A setup that looks efficient in calmer conditions can still be fragile if legal responsibilities, liquidity arrangements, or crisis procedures aren't clear.
The concern isn't new. In 2025, European Central Bank President Christine Lagarde raised similar questions about redemption risks involving stablecoins issued both inside and outside the European Union. What the FSB is doing is putting those concerns into a broader international frame: stablecoin oversight can't be judged only within the jurisdiction where a product is marketed.
What the FSB's AI practices are designed to address
The FSB asked G20 members to support its finalized practices for responsible AI adoption in the financial sector. The guidance covers organization-wide governance, AI development and deployment, cybersecurity, and risks linked to third-party services. The letter also warns that sophisticated cyber threats enabled by AI could undermine the resilience of critical financial institutions and market infrastructure.
This isn't an argument that AI automatically makes finance less stable. Firms can use AI in many ways, and it may improve efficiency or help detect risks. The FSB's concern is about governance - how these systems are managed - how they can fail, and whether common dependencies create vulnerabilities that are hard to see from one institution's perspective.
Governance must cover the whole lifecycle
Approving AI use at a high level isn't enough. Governance has to follow the system from selection and development through testing, deployment, monitoring, and eventual retirement. Firms need to know what a model is being used for, who has authority to deploy it, and what happens when performance changes or outputs can't be trusted.
Controls will differ depending on the use case. A model that sorts internal documents isn't the same risk as a model that affects customer access, credit decisions, trading, or operational controls. Treating every AI system as equally critical can waste resources. Treating everything as low risk because it's "just software" is also a problem. A practical approach is to classify systems based on impact, dependencies, and how much harm they could cause if they fail.
Cybersecurity and third-party exposure
AI can shift the cybersecurity landscape by helping attackers create or adapt threats. The FSB's warning focuses on the potential effect on critical financial institutions and market infrastructure. For firms, the issue isn't only whether they use AI tools. They also need to consider how suppliers, service providers, and counterparties use technology - and how disruptions there could ripple into day-to-day operations.
Third-party concentration is familiar in financial services. Multiple institutions may depend on a small number of cloud, data, or software providers. If an important service is disrupted or compromised, the impact may extend beyond a single firm. AI adoption can deepen these dependencies, especially when critical processes rely on external models or infrastructure that an institution can't easily inspect or replace.
Practical steps include mapping important AI-enabled services and their suppliers, assigning clear accountability for outcomes, testing incident response, and maintaining fallback procedures. These controls don't remove risk - but they make it more likely a firm can detect problems, limit damage, and keep essential operations running.
The FSB had proposed 12 practices for responsible AI adoption in June, then finalized its guidance ahead of the G20 meeting. The takeaway isn't that one checklist settles everything. It's that supervisors are trying to set a common baseline while firms' technology stacks and vendor relationships keep changing.
What G20 policymakers and firms should take from the warning
The G20 meeting gives finance ministers and central bank governors a place to discuss shared priorities. The FSB's letter should be read as a prompt for coordination, not as proof that a new global rule is already in place. The next practical question is how participating authorities translate shared expectations into supervision, regulation, and crisis planning.
For stablecoins, the gap between published rules and real operational readiness deserves attention. A jurisdiction might have a regulatory framework on paper, while cross-border information sharing, recovery planning, or crisis management hasn't been tested. Differences between national regimes don't automatically mean every arrangement is unsafe, though. The right assessment depends on the specific structure - who does what, what assets back it, how redemption works, and which legal responsibilities apply.
For AI, policymakers face a similar implementation problem. Broad principles have to become controls that match the use case and potential impact. If requirements are too vague, firms may struggle to apply them consistently. If controls are too prescriptive, they can become outdated as tools and deployment patterns evolve. The goal is a baseline that supports resilience without assuming one governance model fits every institution.
A practical checklist for financial firms
Firms can use the FSB's concerns to check whether their controls work in practice, not just on paper:
- Map the system. Identify relevant issuers, reserve managers, banks, service providers, AI models, data sources, and operational dependencies.
- Locate the obligations. Track which entity is responsible for redemption, customer outcomes, security, and incident response - especially where responsibilities cross legal entities.
- Test stress scenarios. For stablecoins, focus on concentrated redemptions and reserve liquidity. For AI, test model failure, cyber incidents, and the loss of a critical third-party service.
- Confirm access to information. Check whether executives, compliance teams, and supervisors can get timely, usable information while an incident is unfolding.
- Practice coordination. Run crisis exercises with relevant counterparties and authorities where possible. A contact list isn't the same thing as a response plan.
- Reassess as the system changes. New issuers, reserve arrangements, vendors, models, or jurisdictions can change the risk profile even if the product name stays the same.
This approach matters well beyond large banks. Stablecoin issuers, exchanges, payment companies, custodians, and fintech firms should understand where their operations depend on banks and other external infrastructure. AI vendors serving financial firms should be ready to explain their role, controls, and incident processes to customers who remain accountable for regulated services.
There's a trade-off. More oversight can increase compliance costs and slow some product decisions. But unclear accountability has its own cost - when crisis hits, firms and authorities may find they can't quickly identify who controls key assets, who can act, or how fast the system can recover. Good safeguards focus on where failure can spread, rather than adding process just because it exists.
Conclusion
The FSB's message is that technology risk is often a question of connections. Stablecoins link token holders to issuers, reserve assets, banks, and authorities across borders. AI links financial firms to models, data, cloud platforms, and cybersecurity dependencies. Weakness in one part of either system can matter well beyond the organization where it starts.
For policymakers, the priority is to make cross-border responsibilities and crisis procedures workable - not just to publish principles. For firms, the practical starting point is to map dependencies, test stress scenarios, and clarify who can act when things fail.
The FSB's letter was submitted ahead of the scheduled October 15 G20 finance meeting, and it doesn't, by itself, determine what decisions participants will make. The key will be watching how regulators turn warnings into concrete supervision and coordination. In the meantime, take a close look at your own stablecoin and AI dependencies before a crisis puts them under real pressure.