AI Cyber Threats and Financial System Risks Highlighted by IMF and Policymakers

Heather Stewart argues that world leaders must confront AI-driven cyber threats to protect trust in the financial system. Stewart argues that international cooperation seems unlikely despite the growing risk of digital fraud.

Supermarket chains across France discarded spoiled food after hackers accessed refrigerator controls and raised temperatures to 50C. IMF Managing Director Kristalina Georgieva cited the incident at the fund's annual meetings in Bangkok last week. Georgieva stressed the importance of protection against malicious actors as AI greatly increases the potential for digital fraud. Policymakers have also warned that an eventual crash in AI-related share prices could threaten financial stability.

Georgieva warned that AI-enhanced cybercrime could undermine trust in the financial system altogether. Social media feeds contain widespread advertisements using fake celebrity endorsements to promote cryptocurrencies or gold. AI also allows scammers to generate traditional email-style fraud far more prolifically. Georgieva recalled that many people distrusted the financial system during Georgieva's childhood in Bulgaria and kept cash outside banks. Georgieva warned that more people could adopt that practice unless policymakers act ahead of the threat.

JPMorgan chief executive Jamie Dimon said this week that the bank would not allow digital agents to act for customers, warning of potentially thousands of rogue agents. AI agents, including Meta's Muse, are starting to offer services that use customers' financial details to find higher savings rates and move money accordingly. Financial regulators worry that widespread adoption could produce destabilizing collective movements of deposits and accelerate bank runs during uncertainty. Regulators also question who bears responsibility when an AI agent makes a disastrous or criminal financial decision for a customer.

Stewart argues that responsibility will probably require clarification through case law, which will emerge only after something goes badly wrong. Stewart warns that legal ambiguity over financial obligations can accelerate financial instability. Stewart cites the 2008 financial crisis, when some banks' products had liability structures so complex that the products became effectively impossible to value, unleashing panic.

Bank of England Governor Andrew Bailey, who chairs the Financial Stability Board of global central bankers, spoke alongside Georgieva. Bailey cited news this week that Anthropic had warned AI models might soon be able to replicate themselves in domestic devices. Stewart questions who would bear responsibility and costs if that caused an internet-connected oven or television to explode. Stewart argues that consumers could reasonably reject connected devices in favor of appliances that cannot be hijacked by self-replicating bots. Some supermarkets in France rushed to disconnect refrigerators from the internet this month to avoid the hack.

Stewart argues that limited adoption of connected kettles and refrigerators might impose little loss on society. However, Stewart argues that consumers abandoning banks to hold cash would impose costs on those consumers and the wider economy. Bailey said, "If we don't have trust in money, we don't operate as a society." Senior financial policymakers are deeply concerned about rapidly developing AI interacting with a financial system that extends beyond heavily regulated banks to social media platforms, payment apps and other tools.

Stewart criticizes Donald Trump for making the proposed renaming of AI as "superintelligence" Trump's main contribution to the debate. Federal Reserve Chair Kevin Warsh used Trump's term on stage at the Bangkok meetings this week. All US government agencies have been ordered to use the term.

Trump's superintelligence taskforce, chaired by Director of National Intelligence Jay Clayton, is establishing mechanisms for AI laboratories to report breaches, hacks, jailbreaks and other problems. However, the initiative does not appear to require laboratories to submit models for testing before launch. The UK has pledged to address AI safety through the UK's G20 presidency agenda next year. G7 finance ministers announced in Bangkok that the ministers would conduct simulations to test the global financial system's resilience to cyber-attacks.

Stewart argues that the simulations could identify vulnerabilities. However, Stewart argues that the urgent questions extend well beyond whether criminals can hack banks. Stewart warns that misuse of AI could trigger financial chaos.