Marijuana may account for at least half of the drugs sold on the dark web, local graduate school students found during a recent hackathon.

Students at the Pardee RAND Graduate School discovered while analyzing hundreds of thousands of listings that marijuana was the most-trafficked drug on the dark web, an encrypted part of the internet that is not indexed by conventional search engines. Many of the listings originated from or were sold to users in the United States, said Todd Richmond, director of Pardee’s Tech & Narrative Lab.

Richmond said the data raises questions about how marijuana’s growing legal status — the drug is now fully legal for adults in 11 states, including California, and permitted for medical reasons in 33 states — has affected the black market demand for the drug. The findings also have implications for the effort to halt vaping-related illnesses, as many tainted e-cigarettes originate from the dark web, he added.

“If marijuana is being sold or purchased within the U.S. on the dark web, the question is how has that changed over time with legalization, and how does that point toward a future where we need more legalization or is legalization not solving the problems we thought it was going to,” Richmond said. “We wouldn’t have thought to ask those questions until we experimented with the data.”

Narcotics as a general category represented 60% of the items sold in the dataset the students analyzed, which totaled 400,000 listings across two years, Richmond said. Jewelry, weapons, counterfeit goods, and personal and financial data were also commonly available, he said. The listings originated most frequently from the U.S. and Europe, particularly France, the Netherlands and Belgium.

“Drug posts were spread out all over the U.S. and Europe, but posts for fraud were almost all coming out of the U.S., which surprised me because I thought it would be more international,” Richmond said.

Professor Osonde Osoba, the hackathon’s organizer, said any insights gleaned from the data should be taken with a grain of salt because the dataset did not include listings from Russia. The tools the students used to study the listings are not yet able to parse Cyrillic text.

“Being able to analyze Cyrillic is super relevant for the types of policy questions we’d like to ask about cyber tools for social media bots,” Osoba said. “If Russia turns out to be the main user of those types of tools, it’s a particularly important blindspot.”

Another limitation of the dataset was the ubiquity of slang across the dark web, which thwarted the students’ attempts to use sentiment analysis algorithms on the listings, Richmond said.

“The slang terms used on the dark web often have the opposite meaning to what the algorithms think they mean,” he said. “We’re probably going to train some of these algorithms on slang — which is difficult because it changes rapidly.”

Leave a comment

Your email address will not be published. Required fields are marked *