Bailey Tells G20 That Frontier AI Could Change the Speed, Scale, and Economics of Global Cyber Risk
FSB Chair and Bank of England Governor opens G20 meeting in Asheville with a letter categorizing cyber risks fueled by frontier models as an immediate threat to financial stability.
On August 31, the Financial Stability Board published the annual letter from Chair Andrew Bailey to the finance ministers and central bank governors of the G20, distributed at the opening of the meeting taking place from August 31 to September 1 in Asheville, North Carolina. The document places frontier AI at the top of the list of immediate threats to financial stability, ahead of sovereign debt, private credit, and stretched valuations.
The sentence that underpins the argument is straightforward. According to Bailey, who also heads the Bank of England, 'frontier AI may have the ability to materially alter the speed, scale, and economics of cyber risk, which could undermine market confidence on a systemic scale, particularly due to highly concentrated third-party providers.' The implicit reading is that a coordinated offensive against a handful of hyperscalers or critical SaaS providers generates a correlation risk for the financial system that traditional prudential regulation does not model.
The Risk Ranking Compiled by the FSB
The letter addresses a stack of systemic vulnerabilities. Bailey lists the possibility of disorderly corrections in sovereign debt markets, fragilities in private credit, rising leverage in bond and equity funds, market concentration among major tech firms, and optimism surrounding AI capex. According to Bailey, these forces interact 'in a way that would amplify a future correction.' However, the new point in this edition is the promotion of AI-fueled cyber risk to the category of immediate threat, placing it ahead of the usual macroeconomic topics.
The argument converges with data that regulators have already had on hand. Analysts at Goldman Sachs projected earlier this week that global AI capex is expected to exceed $1 trillion by 2026, up from the previous consensus of $800 billion. According to Mike Mayo, an analyst at Wells Fargo, 'AI is the number one driver' of profits for major U.S. banks this year, a thesis that aligns with the results of the five largest Wall Street firms, which reported $114 billion in capital markets revenue in the first half, up 31.5% year-over-year. It is this same boom that the FSB is now signaling as a potential source of fragility.
Where Local Supervisors Need to Respond
The letter urges national supervisors to treat 'the safe and responsible release and deployment of frontier models' as a global priority and to require financial institutions to have 'robust response and recovery capabilities, as well as resilience among critical third-party providers.' The practical directive is clear: continuity exercises can no longer assume cloud disruptions, enterprise LLMs, and identity providers happen in silos.
In the UK, the Bank of England and the Prudential Regulation Authority have already indicated they will tighten the next round of cyber stress testing to include scenarios involving simultaneous compromises of models used by multiple banks. In the United States, the Federal Reserve and the Office of the Comptroller of the Currency are exploring expanding the Interagency Guidance on Third-Party Relationships to explicitly cover generative AI providers. In Japan, the Financial Services Agency issued guidance on AI usage in July, which is likely to be revised based on the new FSB text. Each supervisor will have to reconcile this requirement with the parallel mandate of not stifling technology adoption.
What Changes in the Bank's Risk Committee
The practice that the CRO needs to institutionalize is the mapping of dependencies on third and fourth parties involving LLMs, cloud providers, and identity providers, with explicit concentration metrics. The regulator will ask. The FSB letter also opens the door for external auditors to start requiring recovery testing with a scenario of combined unavailability of a hyperscaler and primary LLM provider, something that is still rare in a typical continuity plan.
There is an opposing reading that needs to be acknowledged. Analysts from the Institute of International Finance and economists from smaller central banks argue that the FSB has a history of exaggerating emerging risks and that the private sector is already training defensive models capable of offsetting the offensive asymmetry. Bailey acknowledges this thesis by stating that the response 'requires coordination between authorities and the private sector,' but chooses to classify as immediate a risk that much of the market still treats as probabilistic. This divergence in classification is what the next 90 days will test when the G20 must decide whether the text becomes a formal commitment in New Delhi or remains in the footnotes of the statement.