Financial regulators and industry experts alike are paying attention to how the industry is using AI agents for trading activity. 

Some fintechs such as Robinhood have launched agentic AI trading platforms where users can build their own agents and direct them to trade on their behalf, while regulators are asking for tighter oversight on how these agents execute trades.

Bank of England Dep. Gov. Sarah Breeden warned this week at the European Central Bank conference in Sintra, Portugal, that autonomous AI agents could trigger a “market meltdown” if left unregulated, according to Bloomberg’s June 30 reports. 

Unwarranted price action 

If every AI agent takes the same side of the trade, that can not only make markets lopsided, it can also drive unwarranted price action, Izzy Conlin, head of strategy and solutions for global markets at trading platform service provider Tradeweb, told FinAi News. 

Conlin said that the concern among regulators is that if most trading desks pull signals from the same handful of foundation models, they could end up making the same calls at the same time — buying or selling in unison — with no diversity of judgment to cushion a shock. 

AI agents differ from legacy algorithmic trading because they operate in a continuous loop rather than stopping once a trade is executed, which raises the stakes if one goes rogue, she said, adding that given a majority of tools are built on the same few LLMs, one rogue LLM could spark volatility. 

Taming agents 

While that risk is minimal — for now — because very few traders use agentic AI for executions, regulators need to keep such events in mind while shaping new rules, Conlin said. 

“You have to have guardrails in place,” she said, adding that the industry needs to learn to walk before it runs with AI being tasked for research and hidden opportunities. 

“AI is really good at analyzing vast amounts of data quickly and finding patterns that humans might miss,” Chris Ackerson, senior vice president of product development at AI-driven data and tech provider AlphaSense, previously told FinAi News. “We have built tools precisely for that.” 

The industry will move toward using AI for active trading at some point, he said. “Then we need to build kill switches into the models along with high explainability.” 

FIs would also need to custom-build models rather than using something out of the box or general LLM tools, Ackerson said. “They need to train it on their own proprietary data, their own decision-making on what gets traded and what doesn’t to avoid herd behavior among models.” 

Breeden said the Bank of England is working with the Bank for International Settlements and Bundesbank to study whether agents can cause herd behavior. It also is weighing guardrails such as circuit breakers or kill switches that could halt market wide trading if AI models malfunction. 

Breeden compared AI models to teenagers, saying they can behave inconsistently and “differently when you’re watching them,” underscoring why the BoE wants a human accountable for every model in production. 

Robinhood did not respond to FinAi News’ request for comment before publication. 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.