Key Takeaways
- • Federal Reserve Vice Chair Michelle Bowman warned that AI tools deployed in banking systems pose dual risks: the same technologies helping firms defend against threats can be weaponized against them, requiring coordinated regulatory oversight across jurisdictions.
- • A Cambridge Centre for Alternative Finance survey found financial institutions are adopting AI at more than twice the rate of their supervisors, with only 20% of regulators reporting "advanced AI adoption" and 43% of authorities having no plans to collect data on industry AI adoption within the next two years.
- • Anthropic's Mythos AI model has been flagged by cybersecurity experts as presenting significant challenges to banking infrastructure, particularly legacy technology systems, prompting emergency engagement between the Fed, Treasury, and major banks on AI cybersecurity vulnerabilities.
- • Central banks and financial regulators lack empirical data on emerging AI harms in the financial sector; only 24% of surveyed authorities currently collect data on industry AI adoption, creating a critical supervisory blind spot that undermines oversight effectiveness.
- • Bowman called for closer coordination among global regulators, standard-setting bodies, and financial institutions to establish frameworks for monitoring and mitigating AI-related risks before the technology poses systemic threats to financial stability.
Summary
Federal Reserve Vice Chair for Supervision Michelle Bowman issued a stark warning about the regulatory gap emerging as artificial intelligence proliferates across the banking system, highlighting that the same AI tools strengthening institutional defenses can be repurposed as offensive weapons against financial firms. Speaking at an Institute for International Finance event, Bowman emphasized the need for coordinated oversight among regulators, acknowledging that the speed of AI adoption in banking has outpaced regulatory capacity to monitor and manage associated risks.
Research published on April 28 by the Cambridge Centre for Alternative Finance, conducted in collaboration with the Bank for International Settlements, International Monetary Fund, and other multilateral institutions, quantified the supervisory lag. The survey of 350 financial institutions, 140 AI vendors, and 130 central banks and financial authorities across 151 countries revealed that financial firms are adopting AI at more than double the rate of their regulators. Only 20% of regulatory authorities reported "advanced AI adoption," and critically, just 24% of surveyed authorities currently collect data on industry AI adoption. Worse, 43% of regulators have no plans to begin collecting such data within the next two years—a gap the report characterized as an "empirical blind spot" that "may undermine the prevailing optimism on AI."
The urgency of Bowman's call for coordinated action intensified following Anthropic's April release of Mythos, an advanced AI model that cybersecurity experts identified as posing significant challenges to banking infrastructure and legacy systems. The threat prompted emergency meetings between the Federal Reserve, U.S. Treasury, and major banks to assess preparedness. Regulators and standard-setting bodies worldwide have subsequently engaged with financial institutions to evaluate how prepared their systems are for emerging frontier AI models, with particular focus on AI-enabled exploitation of hidden system vulnerabilities.
The regulatory challenge extends beyond detection. Banks are being urged to evaluate AI risk mitigation technologies and services, but regulators themselves lack the data infrastructure and AI capabilities to effectively oversee these deployments at scale. Without coordinated international frameworks and stronger regulatory adoption of AI monitoring tools, Bowman's warnings suggest that financial institutions may face systemic vulnerabilities as AI capabilities continue to advance faster than supervisory capacity to manage them.
Research published on April 28 by the Cambridge Centre for Alternative Finance, conducted in collaboration with the Bank for International Settlements, International Monetary Fund, and other multilateral institutions, quantified the supervisory lag. The survey of 350 financial institutions, 140 AI vendors, and 130 central banks and financial authorities across 151 countries revealed that financial firms are adopting AI at more than double the rate of their regulators. Only 20% of regulatory authorities reported "advanced AI adoption," and critically, just 24% of surveyed authorities currently collect data on industry AI adoption. Worse, 43% of regulators have no plans to begin collecting such data within the next two years—a gap the report characterized as an "empirical blind spot" that "may undermine the prevailing optimism on AI."
The urgency of Bowman's call for coordinated action intensified following Anthropic's April release of Mythos, an advanced AI model that cybersecurity experts identified as posing significant challenges to banking infrastructure and legacy systems. The threat prompted emergency meetings between the Federal Reserve, U.S. Treasury, and major banks to assess preparedness. Regulators and standard-setting bodies worldwide have subsequently engaged with financial institutions to evaluate how prepared their systems are for emerging frontier AI models, with particular focus on AI-enabled exploitation of hidden system vulnerabilities.
The regulatory challenge extends beyond detection. Banks are being urged to evaluate AI risk mitigation technologies and services, but regulators themselves lack the data infrastructure and AI capabilities to effectively oversee these deployments at scale. Without coordinated international frameworks and stronger regulatory adoption of AI monitoring tools, Bowman's warnings suggest that financial institutions may face systemic vulnerabilities as AI capabilities continue to advance faster than supervisory capacity to manage them.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.