The Bank of England has issued one of its starkest warnings yet about the growing role of artificial intelligence in financial markets, cautioning that a wave of similarly-built trading bots could herd into the same positions and turn an ordinary wobble into a synchronised crash. In its latest assessment, the BoE’s Financial Policy Committee (FPC) flagged that as hedge funds and asset managers race to deploy autonomous trading agents, the very efficiency that makes these systems attractive may also create a dangerous uniformity of behaviour that regulators cannot fully observe — let alone control.
The concern is not that any single algorithm misbehaves, but that thousands of them, trained on overlapping data and optimised toward similar objectives, react identically to the same market signals. In a stressed moment, that correlation could collapse the diversity of opinion that ordinarily keeps markets liquid — and replace it with a stampede.
Why Correlated Machines Worry the Watchdogs
Modern financial firms increasingly rely on machine-learning models that ingest news, price feeds and alternative data, then execute trades at speeds no human can match. The FPC’s worry is structural: if many institutions license similar foundation models, fine-tune them on comparable datasets, and reward them with comparable risk-return objectives, their outputs may converge.
“The danger isn’t a rogue bot — it’s a thousand rational bots all reaching the same conclusion at the same millisecond,” said Dr Helena Marsh, a systemic-risk researcher at the fictional Cambridge Institute for Financial Stability. “Diversity of strategy is what makes a market resilient. Homogenised AI quietly erodes that diversity until one day there’s nobody left to take the other side of the trade.”
That dynamic echoes the 2010 ‘flash crash’ and subsequent micro-crashes, where automated systems amplified selling pressure in seconds. But the FPC’s concern goes further: today’s models can adapt, learn and even anticipate each other, raising the spectre of feedback loops that evolve faster than any circuit breaker can respond.
The Observability Problem
Perhaps the thorniest issue is that supervisors cannot easily see inside these systems. Traditional stress tests assume regulators can model how firms will behave under pressure. But autonomous agents may pursue strategies their own operators struggle to explain, let alone predict.
“You can’t stress-test a black box you can’t observe. We’re being asked to model the collective behaviour of systems whose individual logic is opaque and whose interactions are emergent. That’s a fundamentally new kind of risk.”
So said Tomás Rivera, a former markets supervisor now advising the fictional consultancy Northgate Risk Partners. He argues that the existing toolkit — capital buffers, position limits, post-trade reporting — was built for human-paced decision-making and may be poorly suited to machine speed.
The FPC has stopped short of proposing concrete rules, instead signalling that it will deepen its monitoring of AI adoption across the financial system. Insiders suggest the Bank is exploring whether firms should disclose the provenance of the models they deploy, and whether shared dependencies on a handful of AI providers constitute a concentration risk in their own right.
An Arms Race the Industry Won’t Slow
For all the official caution, commercial momentum is firmly in the other direction. Quantitative hedge funds view autonomous agents as the next competitive frontier, and few are willing to unilaterally disarm while rivals press ahead.
- Funds report meaningful gains from agents that trade across asset classes without human sign-off.
- Cloud and AI vendors are aggressively marketing trading-ready models to institutional clients.
- Smaller firms increasingly rent rather than build, deepening reliance on a few shared systems.
“Nobody wants to be the desk that switched off its AI the year everyone else doubled their returns,” noted Priya Anand, a markets strategist at the fictional brokerage Levenmoor Securities. “The incentives push everyone towards the same tools — which is precisely the problem the Bank is describing. Individually rational, collectively fragile.”
That tension — between firm-level advantage and system-level safety — sits at the heart of the regulatory dilemma. Voluntary restraint is unlikely; heavy-handed rules risk pushing activity offshore or into less transparent corners of the market.
What This Means
The Bank of England’s intervention marks a shift from treating AI as an operational curiosity to recognising it as a potential source of systemic risk. For now, the warning is exactly that — a warning, not a rulebook. But it puts hedge funds, asset managers and AI vendors on notice that correlated machine behaviour is firmly on the regulatory radar, and that disclosure requirements or new stress-testing regimes may follow. The deeper challenge is philosophical as much as technical: how to govern a market increasingly steered by systems that move faster than human oversight and think in ways their own creators cannot fully trace. Until regulators solve the observability problem, the FPC’s quiet message to the industry is clear — the next flash crash may not be caused by a glitch, but by everyone’s machines agreeing at once.
Photo by Breakingpic on Pexels
