Two international stock exchange operators have announced plans to launch artificial intelligence tools for market surveillance in the coming months, with a Wall Street regulator not far behind.
Executives are hoping computers with humanoid wit can help mere mortals catch misbehaviour more quickly.
They hope the software will, for instance, scrub social media messages to detect dubious bragging or back slapping around the time of a big trade.
It could also more quickly unravel complex issues, like "layering," where orders are rapidly sent to exchanges and then cancelled to artificially move a stock price.
In the US, the Financial Industry Regulatory Authority (FINRA) hopes AI can help uncover previously unknown forms of market manipulation.
"The biggest concern we have is that there is some manipulative scheme that we are not even aware of," FIRNA executive vice president for market regulation Tom Gira said.
"It seems like these tools have the potential to give us a better window into the market for those types of scenarios."
FINRA plans to test artificial intelligence software being developed in-house for surveillance next year, while Nasdaq and the London Stock Exchange Group expect to use it by year-end.
The exchange operators also plan to sell the technology to banks and fund managers, so that they can monitor their traders.
While financial firms are already applying artificial intelligence software to everything from compliance to stock-picking, it is only starting to become useful for market oversight.
"We haven't really let the machines loose, as it were, on the surveillance side," said Bill Nosal, a Nasdaq business development executive who is overseeing its artificial intelligence effort.
50 billion events
Market surveillance generally relies on algorithms to detect patterns in trading data that may signal manipulation and prompt staff to investigate.
But the sheer volume of data can lead to an overwhelming number of alerts, many of which are false alarms.
FINRA monitors roughly 50 billion market "events" a day, including stock orders, modifications, cancellations and trades. It looks for around 270 patterns to uncover potential rule violations.
It would not say how many events are flagged, or how many of those yield evidence of misbehaviour.
The "machine learning" software it is developing will be able to look beyond those set patterns and understand which situations truly warrant red flags, Gira said.
In the case of market surveillance, machine learning would mean the computers "learn" which trading patterns lead to enforcement charges, in order to flag the right ones.
FINRA plans to test the new tool next year alongside its existing systems to compare the results.
The regulator has already moved its surveillance systems to the Amazon cloud, giving it more computing power to quickly analyse massive amounts of data.
Nasdaq is working with cognitive computing firm Digital Reasoning, which it invested in earlier this year.
LSE has teamed up with IBM Watson business and cyber-security firm SparkCognition to develop its AI-enhanced surveillance, Chris Corrado, chief operating officer of LSE Group said.
The technology would not necessarily prevent events such as the 2010 "flash crash," when the Dow Jones Industrial Average temporarily plunged more than 1000 points.
However, it could be quicker to catch manipulative behaviour thought to contribute to them, potentially saving market watchdogs time and money.
FINRA, Nasdaq and LSE would not provide specific figures for how much the software costs to develop or how much money they expect it to save.
For instance, investigators spent years cross-referencing trading data with old electronic communications to make their case against a group of global banks whose traders were rigging foreign exchange benchmarks.
Nasdaq said the software it is testing with Digital Reasoning and other financial firms could do that task almost in real time.