BIS Warns AI Adoption Could Amplify Risks Across Global Financial Markets

On November 3, Bank for International Settlements official Alexandre Tombini warned that financial institutions using the same AI models and technology providers could make similar decisions during market stress. He also raised concerns that failures at major suppliers could disrupt several institutions and intensify market volatility.

An October 26 survey by the Bank of England and FCA found that external providers supplied 53% of AI applications considered highly material, up from 35% in 2024. Another 54% of firms reported difficulties applying risk management standards to vendor-supplied models.

Banks already use AI in lending, fraud detection and market analysis. As automated decisions become more common, regulators are examining how firms oversee these systems and prevent technology failures from affecting trading and financial services. 

Banks’ AI Dependence Concentrates Technology Risks

An October 26 survey by the Bank of England and FCA identified 169 cloud and computing provider relationships. Of these, 37% were considered critical and difficult to replace. Only 35% of that group had tested plans to leave their providers if necessary. 

Major banks are expanding their technology partnerships. In June, HSBC announced an agreement with Google Cloud to develop more than 200 AI applications over two years.

On October 19, Emirates NBD announced plans to use OpenAI models on UAE-based infrastructure for customer services and internal operations. 

The risks of shared infrastructure became evident on October 20, when a power failure disrupted Google Cloud’s London operations for more than six hours.

Revolut customers experienced problems with payments, withdrawals and account access, while Lloyds was also affected. The outage involved cloud infrastructure, not a reported AI-model failure.

The BIS has warned that outsourcing can weaken banks’ control over critical services. The Financial Stability Board’s October guidance calls for stronger oversight of technology suppliers and business continuity arrangements.

Yet only 27% of firms in the UK survey reported additional resilience measures, including joint incident testing or exit planning, for third-party AI systems. 

AI Trading Models Raise Market Liquidity Concerns

AI use in securities trading remains limited. A February 2026 European Securities and Markets Authority survey recorded 847 AI applications, including 10 for algorithmic trading, three for high-frequency trading and 20 for portfolio risk management. Most applications supported internal operations. 

An April OECD report on Italian financial institutions found that 21% of respondents used AI for predictive analysis, 17% for market analysis and 11% for investment research. These applications are increasingly informing investment decisions, although direct trading remains less common.

Automated trading has previously contributed to market disruption. During the May 2010 Flash Crash, a $4 billion automated futures sell order helped trigger a sharp fall in US markets as liquidity weakened. The SEC and CFTC linked the disruption to conventional trading algorithms, not generative AI.

On October 21, BIS official Tao Zhang warned that institutions using similar AI models could sell assets simultaneously during market stress.

Such activity could deepen price declines, particularly in less liquid markets. The Federal Reserve has also noted that more varied AI trading signals could reduce correlated trading. 

Autonomous AI Raises Financial Transaction Control Risks

AI agents are beginning to execute payments, although their authority remains limited. An October survey by the Bank of England and FCA found that 5% of firms allowed agents to execute payments under set controls, while 2% reported conditional autonomy in payment operations. None reported broad autonomy in payments. 

On October 20, Visa, BILL and Billtrust launched a live pilot allowing AI agents to complete supplier invoice payments. The agents use verified identities and tokenised payment credentials. Buyers must approve payments beforehand, leaving agents to execute instructions rather than decide which invoices to pay. 

The IMF’s April research identified three stages requiring safeguards: payment intent, authorisation and settlement. Federal Reserve Governor Christopher Waller also warned in September that delegated payments raise questions about authentication and liability when an agent makes an incorrect purchase. 

Fraud remains a concern for payment regulators. The European Banking Authority’s June 2026 assessment found that 52% of surveyed banks considered fraud a major operational risk.

A separate EBA–ECB report recorded €4.2 billion in payment fraud across Europe in 2024, although those losses were not attributed to autonomous AI. Wider use of payment agents would require clear records of customer approval and responsibility for disputed transactions.

Financial Regulators Face New AI Oversight Decisions

EU AI Act transparency rules took effect on August 2, 2026. Providers of certain existing AI systems have until December 2 to meet synthetic-content marking requirements. Rules covering prohibited AI practices and general-purpose models are already in force. 

From December 2, 2027, high-risk AI used to assess individual creditworthiness must meet requirements for record-keeping, human oversight and accuracy. Fraud detection is excluded from this specific high-risk category. 

The Digital Operational Resilience Act has applied since January 17, 2025, requiring covered financial firms to manage technology risks, test resilience and report major incidents. European regulators also designated 19 critical technology providers for direct oversight in November 2025. 

The European Central Bank’s 2026–2028 supervisory programme will examine how generative AI affects banks’ risk profiles and governance. Its findings will inform future supervision as institutions decide which automated financial actions require human approval.