Machine Learning Used to Thwart Terrorists
Photo: GettyTerrorists are increasingly using cyber attacks to break into banks and steal money to fund their operations. However, one company said its software can better track and detect these types of intrusions.
“These bad actors are becoming more and more sophisticated and more technologically savvy,” said James Heinzman, executive vice president of financial services solutions at ThetaRay. “They really are trying to exploit the weaknesses in our financial systems.”
ThetaRay — which was established in 2013 and is based in Hod HaSharon, Israel, with a U.S. headquarters in New York City — has developed a program that uses machine learning to detect when a terrorist or hacker breaks into a bank or financial institution, he said.
Such organizations usually store massive amounts of information about financial transactions and users, which make it difficult to sort through, Heinzman said.
“These data lakes become data sewers, because there’s so much data they don’t really know how to analyze it, how to bring it together and find meaningful results,” he said. “It’s like looking for a needle in a needlestack.”
Compounding the issue is the fact that traditionally, such organizations use rules-based algorithms to detect when something is amiss, Heinzman noted.
“Fifteen years ago, that was sort of the cutting edge,” he said. The challenge then “is that in order to find something, in order to find bad behavior, you have to write a rule that specifically addresses it,” he added.
Many of these rules are published on the dark web, which makes it easy for bad actors to breach security systems and wreak havoc undetected, he said. ThetaRay’s approach, however, is different.
“We don’t have any rules, settings, thresholds or parameters,” he said. “Our system automatically tunes to … [various datasets] so it’s able to identify anomalies that are suspicious but difficult to write a rule [for], or that don’t cross a specific threshold or setting. It’s just anomalous behavior within the population.”
Because the system uses unsupervised machine learning, it can comb through piles of information and define what is considered normal behavior or what is unusual, he said.
“We don’t have to know anything about the data,” he said. “We let the data interpret itself.”