Researchers from the Massachusetts Institute of Technology have developed an approach to identify the hidden hands running the illegal, anonymous marketplaces on the darknet.
The nefarious websites, where narcotics and weapons are sold, open and close rapidly to elude authorities, but MIT’s Lincoln Laboratory of Artificial Intelligence has augmented a tool, created by a defense research agency, that has led to increased arrests.
FBI and intelligence officials declined to comment publicly about specific operations, but they privately acknowledge excitement about efforts against the darknet that involve some of the world’s most advanced pattern recognition technology.
“We’re lighting up parts of the internet that exist beyond the reach of mainstream search engines,” a former U.S. intelligence official told The Washington Times on the condition of anonymity.
Major darknet stings in recent weeks include the shuttering of two of the world’s largest illegal drug sites — Wall Street Market and Valhalla — where dealers peddled opioids such as fentanyl, oxycodone and hydrocodone.
In the Wall Street Market crackdown, the FBI and European and South American law enforcement partners worked for nearly two years to “hunt for even the tiniest of breadcrumbs to identify criminals on the darkweb,” U.S. Attorney McGregor W. Scott said.
FBI and Europol operatives notched another major victory last week when they dismantled DeepDotWeb, which functioned as a portal between the darknet and the internet. Criminals used DeepDotWeb to launder millions of dollars in illicit funds, including the cryptocurrency bitcoin, officials said.
At the core of these operations, analysts say, is a revolution in understanding the movement of criminals across the darknet, a landscape that traditionally has been almost impossible for police to navigate because of its “layered, onionlike nature.”
“The darknet is aptly named ’the onion’ because it exists layer upon layer, and every time law enforcement pulls back a layer, there are a hundred more,” said Mark Lanterman, a cyberforensics specialist and chief technology officer at Computer Forensic Services based in Minnetonka, Minnesota.
Charlie Dagli, a researcher at the MIT Lincoln Laboratory’s Artificial Intelligence Technology and Systems Group, said the shape-shifting nature of darknet operators poses unique challenges, with sites closing constantly because they are raided, hacked or abandoned.
“This constant switching between sites is now an established part of how dark[net] marketplaces operate,” Mr. Dagli told MIT News this week.
There are also what authorities call “exit scams” in which a site closes immediately after customers pay for unfulfilled orders, scamming users out of their payments.
A key in the Wall Street Market investigation was a bitcoin exit scam “believed by investigators to be [worth] approximately $11 million,” according to Department of Justice court documents.
But improvements have aided policing efforts, with some of the more groundbreaking work at MIT’s Lincoln Laboratory. Efforts are underway to develop artificial intelligence tools that evolved from a Defense Advanced Research Projects Agency (DARPA) project called MEMEX.
MEMEX ran from 2014 to 2017 and included dozens of universities, national laboratories and companies pioneering a darknet search engine that involved text, speech and visual analytics. DARPA ultimately released the project as open-source software.
Manhattan District Attorney Cyrus Vance Jr. said MEMEX helped his office screen more than 6,000 arrests for signs of human trafficking in 2017.
It also improved New York Police Department anti-prostitution work, with MEMEX helping increase related online arrests from 15 to 300 that year.
MIT researchers now are working to recognize patterns in darknet behavior, including developing algorithms that monitor how users identify one another and what they write about.
Mr. Daglia said data quantity remains a major issue. Any given investigation could involve sifting through millions of online sex ads or hundreds of thousands of posted cellphone numbers.
Machine learning, however, is allowing for improved data analysis that also detects sophisticated nuances along the way.
An example cited by MIT involved an algorithm successfully tracking users switching names from “sergeygork” to “sergey gorkin” while moving around different sites. More subtle changes such as “joe knight” to “joe nightmare” also were detected, in addition to similarities in what content was shared.
“Every time we report a match, we are correct 95% of the time,” Mr. Dagli was quoted as saying.
Mr. Lanterman told The Times that he remains skeptical and said many machine learning programs are starting to come under great scrutiny in courts.
“AI is still in its infancy, especially in law enforcement,” he said, adding that “the government is making strides at making people less invisible but then all it takes is some programmer to make them invisible again.”
• Dan Boylan can be reached at dboylan@washingtontimes.com.
Please read our comment policy before commenting.