How fat are AI trading bots’ fingers?

{"text":[[{"start":5.8,"text":"Trista Kelley is a freelance writer."}],[{"start":8.3,"text":"The Bank of England and Financial Conduct Authority published their latest survey of AI in UK financial services at the end of October, and the gist is that while everyone is amping up their use of agents, no one’s letting them off the leash. Of the survey respondents, 80 per cent have AI agents deployed, but only 2 per cent have any “broad autonomy”. Fewer than one in 10 expect agents to execute trades within the next year."}],[{"start":33.16,"text":"As the BoE puts it, “AI agent deployment is widespread, but autonomy is limited”."}],[{"start":39.14,"text":"While the survey’s sample size was small (it’s a self-reported survey of UK-regulated firms, and only 100 or so completed the full questionnaire), the messaging tracks with what big players are saying, too."}],[{"start":50.82,"text":"Giants like JPMorgan, Goldman Sachs, and hedge funds Citadel and Bridgewater, all say their AI agents are barred from making their own calls on trades. So firms, rightly, are worried about rogue Skynet traders."}],[{"start":64.22,"text":"What they aren’t worried about, though, is telling. See below:"}],[{"start":null,"text":"

"}],[{"start":68.18,"text":"Consumer and market conduct — regulatorese for things like treating customers fairly and market manipulation — is literally the least of executives’ worries. It’s also one of the areas where firms say they are least likely to have mitigation controls."}],[{"start":81.86,"text":"Perhaps firms are too confident that keeping AI away from the trading desk is enough to protect market integrity. One Skynet-esque omen: Jane Street’s co-head of technology, Ron Minsky, recently described a “serious security problem” in which AI agents tried to work around their own permissions."}],[{"start":99.34,"text":"And a Wharton School study, revised last year, analysed what could happen if machines are given free rein to trade against other machines. The authors say AI trading bots can end up colluding to manipulate prices “without any form of agreement, communication, or even intent”."}],[{"start":114.78,"text":"Such collusion compromises market efficiency, the authors note, “by decreasing liquidity, diminishing price informativeness, and widening mispricing.” Those sound like ideal conditions for a flash crash."}],[{"start":126.64,"text":"But even if firms keep AI traders away from the P&L, vulnerabilities that could trigger a meltdown still lurk in the market plumbing."}],[{"start":133.5,"text":"Industry execs are already fretting about this. The BoE survey highlights that they rank cyber risk amplification and third-party dependencies as the highest systemic dangers on their radar today and over the next three years. And they see interconnectedness (automated trading, common failure modes) as the fastest-growing risk."}],[{"start":152.68,"text":"The BoE flagged a related risk in July. In its Financial Stability Report, the bank found that while frontier models are increasingly good at finding and exploiting software flaws, the scramble to patch them could itself disrupt tangled, interconnected networks."}],[{"start":167.06,"text":"As Bank researchers warned:"}],[{"start":168.74,"text":"A materially higher volume of identified vulnerabilities would require firms and suppliers to patch and validate changes at much greater speed and frequency, increasing the risk of errors, outages, and disruption across interconnected systems."}],[{"start":182.54,"text":"The October survey hints at how fast things could unravel. Third-party vendors now supply 53 per cent of the AI systems that firms consider most important to their business, while three providers account for 79 per cent of the cloud and compute relationships that firms reported. Only 24 per cent claim deep knowledge of third-party systems, while 81 per cent lean on vendor assurance."}],[{"start":null,"text":""}],[{"start":204.46,"text":"So the AI-era blow-up might not look that different on the outside from a regular fat finger. The old market blunder was caused by a distracted, hungover or merely unlucky trader on a trading floor. Today’s market crash could be caused by a distracted, hungover or merely unlucky vendor pushing out a broken update at a tech firm."}],[{"start":222.98,"text":"And instead of a trade going wrong on one desk, dozens, even hundreds, of trading desks could miscalculate risk at the exact same time."}],[{"start":230.82,"text":"Good luck with that, human traders."}],[{"start":235.94,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1794583129_3761.mp3"}