How can we address the financial risks of AI?

As the world leaders gathered for the UN General Assembly in New York, the question of whether artificial intelligence (AI) could trigger a financial crisis remained a topic of concern for some investors and data center debt issuers. However, another pressing issue emerged: what if malicious humans used AI to launch devastating cyberattacks against key infrastructures such as financial markets?

Bill Gates, co-founder of Microsoft and the Bill and Melinda Gates Foundation, warned that AI was the most dangerous technology ever created, urging governments and tech companies to implement surveillance systems to detect and prevent malicious use.

China and the United States are even discussing setting up a common hotline to discuss AI risks. While these discussions are a step in the right direction, much more needs to be done. While political leaders, investors, and voters assimilate ambitious promises, they must also pay attention to the less glamorous arena of global regulation. Between supervising bodies, debates are emerging on how to address financial risks derived from AI, ranging from global high-level agreements to the detailed use of autonomous AI systems.

In a meeting on digital regulation held in Cambridge earlier this month, overseen by Agustín Carstens (former IPIB Director) and Ashley Alder (President of the UK Financial Behaviour Authority), nearly 300 teams presented proposals for autonomous regulator hackathons. Three key lessons emerged from this intense event. First, regulators recognize that they cannot wait for politicians to create adequate laws or regulatory structures to control AI; the sector moves faster than any legislature can act or reach diplomatic agreements.

As one high regulator stated, "We need to be very creative with the powers we already have in regulation. It will not be easy, as bureaucrats tend to be conservative by culture." However, it is essential; as Carstens pointed out, most financial AI risks are outside their regulatory scope, particularly due to banks' excessive concentration on cloud service providers.

Second, perhaps the concept of explainability needs to be rethought. Regulators have long demanded transparency, interpretability, and accountability in the financial sector, but this is difficult to enforce for AI. Some might consider this a reason to prohibit AI, but such an attitude seems unrealistic. Main regulatory bodies are discreetly adopting a results-oriented approach, implementing safeguards such as emergency disconnection mechanisms while promoting greater clarity on legal responsibility.

Lastly, regulators must act promptly by creating their own AI tools to understand the technology and improve supervision. Some are already doing so, like the impressive variety of AI agents exhibited at the Cambridge hackathon, ranging from capital flow trackers to cyberattack detection agents. The most striking aspect is that many of the most creative ideas emerged from non-Western countries such as India, Pakistan, Rwanda, Kenya, and the United Arab Emirates, where autonomous AI systems will manage half of government services by mid-2028.

It is unlikely these experiments will garner much attention or reassure AI advocates demanding drastic measures to curb technological progress.