{"text":[[{"start":5.68,"text":"Could AI spark a financial crisis? That was one question hanging over New York this week, as world leaders gathered for the UN General Assembly."}],[{"start":14.72,"text":"For while some investors are fretting that the AI stock bubble might burst and/or some data centre debt go into default, many UNGA attendees raised another question: will malevolent humans use AI to launch devastating cyber attacks on key infrastructure — such as financial markets?"}],[{"start":32.84,"text":"Indeed, Bill Gates, founder of Microsoft and the Gates Foundation, thinks that such threats are so real that he told me that AI was now “by far” the most dangerous technology ever created. He wants governments and tech companies to install AI monitoring systems to indicate (and hopefully prevent) any attempted malevolent use."}],[{"start":53.42,"text":"Maybe this will happen; the news that China and the US are considering a joint “hotline” to discuss AI risks is a small symbolic step in the right direction. So is the prevalence of such debates at UNGA. Drastically more needs to be done, however. And as politicians, investors and voters digest these lofty pledges, they should also track another less glamorous — and less visible — realm: global regulation."}],[{"start":78.94,"text":"Out of public sight, debates are now bubbling among supervisors about how to tackle financial risks from AI. This is not just when it comes to top-level global deals but by using agentic AI systems in a granular way themselves. In plain English, some bureaucrats are now trying to fight digital fire with fire."}],[{"start":96.72,"text":"This might sound counterintuitive. But I saw one example earlier this month at a meeting on digital regulation in Cambridge, hosted by an initiative of which I am an adviser. It was co-convened by Agustín Carstens (former head of the Bank for International Settlements) and Ashley Alder (chair of the Financial Conduct Authority), and featured what was billed as the first global “Agentic Regulator” hackathon."}],[{"start":119.64,"text":"Dozens of western and non-western regulators, from both financial and non-financial spheres, participated, while nearly 300 teams made submissions. From this frenzy, three key lessons emerged."}],[{"start":131.54,"text":"First: supervisors now realise they cannot afford to wait for politicians to create the right laws or regulatory structures to control AI; the field is moving faster than any legislature can act or cut diplomatic deals. As one top regulator observed: “We have to be very creative with the powers we have on the books.”"}],[{"start":153.32,"text":"That will not be easy; bureaucrats are culturally conservative. But it is essential: as Carstens has noted, one massive headache for central banks is that most AI financial threats lie outside their regulatory perimeter. Just look at the excessively concentrated exposure of banks to cloud providers."}],[{"start":172.8,"text":"In America, Scott Bessent, the Treasury secretary, is tackling this problem from his own bully pulpit; he recently summoned bankers to discuss Anthropic’s Claude Mythos AI model and concerns that it might identify financial cyber vulnerabilities. In the UK, the Bank of England has just pledged to supervise cloud companies as “critical third parties”. The BIS is running its own innovation experiments. But we need to break regulatory silos."}],[{"start":200.76,"text":"Second: the concept of “explainability”, or the degree to which digital models are transparent, interpretable and accountable, might need to be rethought. Regulators have long demanded explainability in finance. But this is hard to enforce in the AI realm."}],[{"start":216.4,"text":"Some might view this as reason for a ban. But that seems unrealistic. So big regulators are quietly shifting to a stance where they focus on tracking AI model outcomes, and install guardrails (like kill switches) while pushing for clarity around legal liability."}],[{"start":232.16,"text":"Third: regulators urgently need to get their hands dirty by creating their own AI tools, both to understand the technology and to boost monitoring. The good news is that some are already doing this: at this month’s hackathon there was a dazzling array of AI agents on display. Some track capital flows. Others monitor market herding, fraud risks, vulnerabilities, financial robo-advisers and so on."}],[{"start":256.6,"text":"And what was striking is that many of the most creative ideas emanated not from western institutions but places such as India, Pakistan, Rwanda, Kenya or the United Arab Emirates (where half of government services will supposedly use agentic AI by mid-2028)."}],[{"start":272.96,"text":"These experiments are unlikely to ever grab media headlines, let alone reassure AI “doomers” who want more dramatic action to slow the technology down. Fair enough: the US and Chinese governments must certainly collaborate to contain the risks in a top-level way and should learn lessons from other global regulatory initiatives around finance or nuclear weapons."}],[{"start":294.52,"text":"But until then, investors and voters should also cheer on those regulators who are quietly trying to fight AI with its own tools. If nothing else, it shows that AI innovation is not the exclusive preserve of private sector entrepreneurs in Shenzhen or Silicon Valley; some government bureaucrats are also now trying to embrace it for better supervision. Big Tech should back that to the hilt."}],[{"start":320.16,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1790314356_2734.mp3"}