Study exposes differential treatment between cannabis and alcohol in Snapchat's artificial intelligence
A study funded in the USA reveals that Snapchat's AI blocks responses about marijuana but provides information about alcohol and delta-8 THC, raising questions about safety and regulation
Published on 10/03/2025

Research indicates that Snapchat's artificial intelligence treats marijuana, alcohol, and delta-8 THC unequally, sparking debate on information access
A recent study funded by a federal agency in the United States identified an apparent discrepancy in the behavior of Snapchat's artificial intelligence (AI): while it provides responses about alcohol and the cannabinoid delta-8 THC, it blocks responses related to traditional cannabis (cannabis with delta-9 THC).
From August 2024 to January 2025, researchers from George Washington University and the University of North Texas conducted tests with Snapchat's AI on accounts simulating users of different ages (adults and 13-year-olds). They noticed that when searching for "marijuana" or "THC," the AI refused to respond, claiming that "drug use can be harmful to health."
Read Also:
Google opens space for cannabis ads in Canada
Anvisa prohibits cannabis products; companies involved claim outdated information
On the other hand, questions about alcohol or delta-8 THC, the study reported that the AI provided information, including sales locations, without presenting scientific references.
What the study says: discrepancy between marijuana and delta-8
The study shows that Snapchat allows content about delta-8 THC even when it blocks any direct mention of marijuana. The authors of the study suggest that this difference may convey to users the idea that delta-8 is safer or more socially acceptable, which could reduce risk perceptions.
Furthermore, the AI did not provide citations or sources for the statements it made about delta-8, raising concerns about the reliability of the information offered to young people. According to the study, this could influence user behavior and have policy implications on how this type of content should be regulated.