{"text":[[{"start":10.95,"text":"Europe’s top banking watchdogs have warned that frontier AI models such as Anthropic’s Mythos could create “systemic risks to the financial system” as they gave lenders a tight four-month deadline to prepare for greater cyber security threats."}],[{"start":26.45,"text":"The European Central Bank’s chief financial supervisor Claudia Buch on Tuesday sent a letter to 110 lenders giving them until the end of October to come up with a “comprehensive action plan” to combat the cyber risks posed by state of the art AI models. "}],[{"start":44.3,"text":"In a separate warning, the European Systemic Risk Board said modern AI models could undermine financial stability across the Eurozone as hackers can use them to cheaply automate highly complex attacks, while banks remain bogged down by regulatory requirements, legacy IT systems and slow updates from software providers."}],[{"start":64,"text":"The concerns were echoed on Tuesday by the Bank of England, which warned that rapid advances in frontier AI capabilities had “increased financial stability risks related to cyber and operational resilience”. "}],[{"start":76.75,"text":"However, BoE governor Andrew Bailey also took a swipe at his EU counterparts, saying: “I saw what the ECB issued today. I think it is sensible in terms of highlighting the issue. But in all honesty we are taking a different approach. We are working very closely with the banks, because . . . it is not about issuing edicts.”"}],[{"start":95.95,"text":"Global regulators have been alarmed by the arrival of a new generation of AI models, which are capable of finding huge numbers of previously hidden IT weaknesses."}],[{"start":105.65,"text":"The ESRB, which is responsible for monitoring and preventing threats to the European financial system, said the latest AI models represent “a paradigm shift for cyber security” as IT weaknesses could now be “weaponised” in a “matter of minutes or hours”. "}],[{"start":120.80000000000001,"text":"The latest technology can discover and exploit bugs in IT systems “at a speed, scale and level of accuracy far exceeding previous AI models”, the watchdog added, as it raised its assessment of systemic cyber risk to “severe” from “elevated”. "}],[{"start":137.20000000000002,"text":"Anthropic, whose models have helped establish the start-up as a leading AI developer alongside rivals including OpenAI, was the first AI company to face White House export restrictions after US officials raised concerns that its technology could be used to exploit cyber security vulnerabilities in critical infrastructure. Those restrictions on its leading models were lifted last week after the company agreed to additional safeguards. "}],[{"start":163.15,"text":"Bailey urged the US to co-operate with other countries on AI. “These models are a big step forward in terms of capabilities and the threat issue is a really major step forward,” he said. “No single country can think that it can solve that resilience problem on its own.”"}],[{"start":179.4,"text":"The EU banking watchdogs did not name Mythos in their warnings but referred to “frontier artificial intelligence models” in general."}],[{"start":187.20000000000002,"text":"The warnings came just hours before the European Commission presented its own action plan to address the risks of advanced AI models for cyber security. "}],[{"start":197.25000000000003,"text":"The commission intends to set up an EU “evaluation capacity” for advanced AI models next year, which will also examine cyber security. Brussels will also better support governments and businesses to ensure the bloc’s critical infrastructure is protected against cyber security vulnerabilities. "}],[{"start":215.65000000000003,"text":"In the meantime, the ECB and ESRB are urging banks to rapidly beef up their IT security capabilities. "}],[{"start":223.25000000000003,"text":"In her letter to Eurozone banks, Buch said lenders’ action plans should include “concrete measures to strengthen relevant controls, allocating the necessary resources, assigning clear roles and responsibilities and defining timelines for implementation”."}],[{"start":237.35000000000002,"text":"Over the short term, lenders needed to be in a position to detect and fend off large-scale attacks quickly and to ensure that their third-party IT service providers are also able to do so, Buch stressed."}],[{"start":248.75000000000003,"text":"In an attempt to compensate for the tight timeline, the ECB is postponing the deadline for its annual IT risk questionnaire, which banks now have until February to file rather than September. "}],[{"start":260.5,"text":"Additional reporting by Barbara Moens in Brussels"}],[{"start":271.34999999999997,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1783477257_4053.mp3"}