As concerns over the potential risks posed by artificial intelligence (AI) to the global financial system grow, policymakers and regulators are increasingly focusing on ways to mitigate these threats. The issue gained prominence during this week’s United Nations General Assembly in New York, where leaders discussed the dual challenges of AI’s economic implications and its potential use in cyberattacks targeting critical infrastructure, including financial markets.

Bill Gates, co-founder of Microsoft and the Gates Foundation, described AI as “by far” the most dangerous technology ever created, emphasizing the need for governments and technology companies to establish monitoring systems aimed at preventing malicious uses of AI. In a rare move toward international cooperation, China and the United States are reportedly considering a joint "hotline" to share concerns related to AI risks, reflecting growing awareness of the global dimension of the issue.

Beneath the high-profile diplomatic discussions, regulators worldwide are engaging in technical efforts to address AI-related financial risks. Recent meetings among supervisory authorities have highlighted the urgency of developing innovative regulatory approaches. One such event, a hackathon focusing on “Agentic Regulators,” was organized earlier this month by former Bank for International Settlements (BIS) head Agustin Carstens and Financial Conduct Authority chair Ashley Alder. The competition attracted nearly 300 teams from financial and non-financial regulatory bodies across multiple countries, including non-Western nations such as India, Pakistan, Rwanda, Kenya, and the United Arab Emirates.

The hackathon underscored several key lessons. First, regulators recognize that legislative bodies often lag behind the rapid pace of AI development, necessitating the use of existing regulatory powers in creative ways to address emerging threats. Carstens has pointed out that many AI-related financial vulnerabilities fall outside traditional regulatory frameworks, such as banks’ heavy reliance on cloud service providers—a challenge currently being addressed by the Bank of England’s commitment to supervise these companies as critical third parties. In the United States, Treasury Secretary Janet Yellen recently convened banking leaders to discuss concerns related to Anthropic’s Claude Mythos AI model and its potential to expose cyber vulnerabilities.

Second, the traditional regulatory demand for algorithmic “explainability”—the ability to interpret and understand decision-making processes in financial models—is proving difficult to enforce in AI systems. While some advocate for outright bans on opaque AI models, most regulators are pivoting toward monitoring AI outcomes and instituting safeguards like kill switches, alongside clarifying legal responsibilities tied to AI applications.

Third, regulators are increasingly developing their own AI tools to enhance oversight capabilities. These internal AI systems are being designed to track capital flows, detect market manipulation, assess fraud risks, and monitor the behavior of automated financial advisors. Notably, some of the most innovative regulatory ideas are emerging from countries outside the traditional financial powerhouses, demonstrating a global effort to harness AI for supervision and risk management.

While these efforts may not offer immediate reassurance to those calling for stronger restrictions or slower AI development, experts stress the importance of collaboration among major powers, particularly the United States and China, to manage AI risks at the highest levels. Meanwhile, regulatory bodies’ growing engagement with AI technologies represents a significant step in integrating advanced tools into financial oversight, signaling a recognition that AI innovation extends beyond private sector development hubs to include public sector actors dedicated to maintaining stability.