Financial regulators rarely speak in unison. When the Financial Stability Board and the Federal Reserve both publish major reports on the same threat within months of each other, the signal is clear: artificial intelligence has moved from experimental tool to systemic risk factor in global finance. [1] [2]
The FSB’s November 2024 report to the G20 revisits its 2017 assessment and finds a transformed landscape. Technological advances and cheaper compute have accelerated AI adoption across financial firms and supervisors alike. Benefits are real — operational efficiency, regulatory compliance, personalized products, and advanced analytics — but the board warns that vulnerabilities now “stand out for their potential to increase systemic risk.” [1]
Four channels that could amplify systemic risk
The FSB identifies four interconnected vulnerability channels. First, third-party dependencies and service provider concentration. A handful of cloud and AI infrastructure providers now underpin critical financial functions. An outage or adversarial attack on one could cascade across institutions simultaneously. [1]
Second, market correlations. When multiple firms use similar models trained on overlapping data, their trading and risk decisions converge. This herd behavior can amplify price moves and reduce market liquidity during stress. [1]
Third, cyber risks. AI expands the attack surface — model poisoning, prompt injection, and data exfiltration join traditional threats. Generative AI also lowers the barrier for sophisticated phishing, deepfakes, and social engineering at scale. [1]
Fourth, model risk, data quality, and governance. Traditional model validation assumes static relationships. AI systems that continuously learn or operate as black boxes defy existing governance frameworks. Poor data provenance or undetected drift can produce decisions that are wrong, biased, or illegal — and difficult to trace. [1]
Generative AI adds fraud, disinformation, and misalignment
The Federal Reserve’s working paper on generative AI underscores a newer threat layer. GenAI “increases the potential for financial fraud and disinformation in financial markets,” the FSB notes, echoing the Fed’s focus on how synthetic content could manipulate prices, trigger runs, or undermine trust in market infrastructure. [1]
Misaligned AI systems — those not calibrated to operate within legal, regulatory, and ethical boundaries — can engage in behavior that harms financial stability. This is not science fiction. An autonomous trading agent optimizing for a narrow objective could exploit market microstructure in ways that violate rules or destabilize venues, especially if guardrails are absent or poorly specified. [1]
Longer-term structural shifts
Beyond immediate vulnerabilities, the FSB flags three structural trends. Market structure may concentrate further as only large firms afford frontier AI capabilities. Macroeconomic conditions could shift if AI-driven productivity gains or job displacement alter credit cycles. Energy use from training and inference at scale may affect energy markets and, by extension, financial exposures to energy-intensive sectors. [1]
What authorities are being asked to do
The FSB does not call for new AI-specific regulation — yet. Instead, it urges national authorities and international bodies to: close information gaps on AI usage across the financial system; assess whether existing policy frameworks remain adequate; and build supervisory capabilities, including by deploying AI-powered monitoring tools themselves. [1]
This pragmatic sequence — monitor, assess, then act — reflects a recognition that heavy-handed rules could stifle beneficial innovation while gaps in visibility leave blind spots. The FSB also notes an OECD-FSB roundtable where public and private experts discussed “opportunities, risks and the role of policymakers in promoting the safe use of AI in finance.” [1]
Practical takeaways for firms and oversight teams
Financial institutions should expect supervisors to ask harder questions about AI inventory, third-party mappings, and model governance. Firms that cannot explain how their AI systems make decisions, where training data originates, or what happens when a key provider fails will face regulatory friction.
Risk managers should stress-test for correlated model behavior across peers, not just idiosyncratic failures. Cyber teams need threat models that include prompt injection and synthetic media. Compliance functions must extend governance to cover continuous learning systems, not just static models.
Regulators, meanwhile, are building their own AI tooling to detect anomalies across markets — turning the technology inward to police its own externalities. The race is not just between firms adopting AI, but between risk emergence and supervisory visibility.
The message from both the FSB and the Fed is consistent: AI in finance is no longer a pilot project. It is infrastructure. And infrastructure-grade risks demand infrastructure-grade oversight.
